The new framework, called AMS-LPP, is described in the Journal of Big Data by Murad Ali Khan, Muhammad Faseeh, Syed Shehryar Ali Naqvi, and Do-Hyeun Kim. Its central insight is deceptively simple: no single machine learning approach works best across all situations. When a new prediction task closely resembles previous ones, a pretrained model can be fine-tuned efficiently. When the task is genuinely different, starting fresh with automated machine learning (AutoML) generalizes better. AMS-LPP makes that choice dynamically, using a statistical measure of similarity between datasets to decide which path to take.

The similarity gauge is the Mahalanobis distance, a metric that accounts for correlations between features when measuring how far one dataset’s statistical distribution lies from another’s. When the distance is small — meaning the new material’s data looks statistically like data the model has already seen — AMS-LPP fine-tunes a pretrained TabNet model, a deep learning architecture designed for tabular data, using Bayesian hyperparameter optimization to squeeze out the best settings. When the distance is large, the framework abandons transfer learning and invokes AutoML pipelines instead, letting automated search build a model from scratch that is robust to the domain shift.

The results are striking. Under high and moderate similarity, transfer learning achieved a coefficient of determination (R²) above 0.99 while cutting training time by up to 35 percent compared with training from scratch. Under low similarity, AutoML took the crown, reaching R² = 0.998 — an almost perfect fit — on the most dissimilar domains. In practical terms, this means the framework adapts its approach the way an experienced human modeler would, but does so automatically and consistently.

Lightfastness prediction is a genuinely hard problem for machine learning. The property depends on complex interactions between dye chemistry, fiber structure, and optical characteristics, and datasets collected across different laboratories, material types, and measurement conditions are riddled with domain shifts — statistical discrepancies that degrade models trained on one set when applied to another. Conventional experimental determination remains costly and slow, and naive machine learning alternatives often collapse when confronted with heterogeneous data.

Transfer learning alone has a well-known weakness: it helps only when source and target domains are similar. AutoML, by contrast, ignores inter-domain relationships entirely, rebuilding models from zero each time. Neither approach alone adapts to varying domain similarity. AMS-LPP resolves this tension by quantifying similarity first and switching strategies second, ensuring that the efficiency of transfer learning is harvested when it is safe to do so and that the generalization strength of AutoML is deployed when it is not.

Crucially, the team did not treat the models as black boxes. Using SHAP (SHapley Additive exPlanations) analysis, they showed that the predictions are driven by physically meaningful optical and color-related features — the very properties that govern how fabrics fade in real life. This explainability matters for scientific acceptance: a model that predicts lightfastness for reasons that align with the physics of photochemical fading is far more trustworthy than one whose reasoning is opaque.

The implications reach beyond textiles. The framework’s architecture — measure domain similarity, then choose between fine-tuning and from-scratch automated modeling — is a general pattern for any materials informatics problem where data arrives from heterogeneous sources. Accurate property prediction can accelerate materials discovery, and the authors argue that AMS-LPP offers a scalable, efficient, and interpretable solution for similarity-aware model selection in material property prediction broadly.

The work was supported by the National Research Foundation of Korea and the Korea Institute of Energy Technology Evaluation and Planning. As datasets in materials science continue to grow and fragment across laboratories and material classes, frameworks that can automatically judge when to reuse knowledge and when to start fresh may become as essential to materials discovery as the experiments they are designed to replace.

Lightfastness itself is a standardized property in textile science, typically graded on blue wool scales or gray scales after controlled exposure to artificial light sources in weathering chambers. Because the grading process is inherently comparative and partly subjective, it has historically resisted automation, and the variability introduced by different dye classes, fiber chemistries, and finishing treatments makes any single predictive model fragile. The Jeju National University team’s decision to frame the problem as a multi-domain challenge, rather than a single regression task, reflects this underlying heterogeneity of real-world textile data.

The choice of TabNet as the transfer learning backbone is notable in this context. Unlike conventional deep networks designed for images or sequences, TabNet uses sequential attention to select which features to reason about at each decision step, making it well suited to the tabular measurement data that dominates materials informatics. Feature selection of this kind also pairs naturally with the SHAP analysis reported in the study, since both approaches operate on the level of individual input features, allowing the researchers to verify that the same optical and color-related variables the model exploits are the ones chemists would expect to govern fading behavior.

The Mahalanobis distance plays a similarly principled role. Unlike simpler metrics such as Euclidean distance, it rescales each feature by its variance and accounts for correlations among features, so a dataset that differs only along a well-characterized, low-variance direction can still register as distant. In domain adaptation settings this matters because what looks like a small shift in raw feature space may in fact represent a substantial change in the underlying data-generating process. By gating the transfer learning decision on this measure, AMS-LPP encodes a rule of thumb that experienced practitioners follow informally — reuse knowledge only when the new problem genuinely resembles the old one — into an explicit, reproducible criterion.

The Bayesian hyperparameter optimization component also deserves attention. Fine-tuning a pretrained network involves choices about learning rates, layer freezing, and regularization that interact in non-obvious ways, and grid or random search can waste considerable compute exploring poor regions of the configuration space. Sequential model-based optimization, the family of techniques behind Bayesian HPO, builds a probabilistic surrogate of the performance landscape and samples promising configurations preferentially, which is one reason the reported 35 percent reduction in training time under high and moderate similarity is plausible rather than optimistic.

The evaluation metrics reported in the study — R² alongside mean absolute error, mean squared error, root mean squared error, and mean absolute percentage error — provide complementary views of model behavior. R² captures the proportion of variance explained but can mask systematic biases at particular ranges of the lightfastness scale, while the error metrics reveal the magnitude of typical mistakes in units that matter for practical grading decisions. Reporting the full suite suggests the authors were attentive to the possibility that a near-perfect R² could coexist with meaningful errors in specific prediction regimes.

The funding context, including support from the National Research Foundation of Korea and the Korea Institute of Energy Technology Evaluation and Planning, situates the work within a broader Korean research push into energy and industrial applications of artificial intelligence. Textile manufacturing remains a significant industry in the region, and property prediction tools that reduce the need for physical weathering trials could shorten development cycles for dyed and finished fabrics considerably.

Methodologically, the study sits within the emerging field of materials informatics, which applies data-driven techniques to accelerate the discovery and optimization of materials. A recurring lesson in this field is that the bottleneck is often not model capacity but data logistics: datasets arrive from different instruments, laboratories, and material systems, each with its own biases and measurement conventions. Frameworks that explicitly model these domain relationships, rather than assuming data are drawn from a single distribution, address a central pain point. The adaptive selection strategy demonstrated here — switching between transfer learning and AutoML based on quantified similarity — offers a template that could be applied to other properties where heterogeneous data are the norm, from corrosion resistance to thermal stability.

The explainability findings carry particular weight for industrial adoption. Regulatory and quality-control environments in textiles often require that decisions be auditable, and a model whose feature attributions align with known photochemistry — the roles of dye chromophores, fiber morphology, and optical scattering in fading — can be more readily integrated into existing certification workflows than an opaque predictor, however accurate. The alignment between SHAP-derived feature importance and physically meaningful optical and color-related variables suggests the models learned genuine structure rather than dataset artifacts, which is precisely the kind of evidence needed to trust predictions on materials not represented in the training data.

Looking forward, the framework’s reliance on a distance threshold to trigger the switch between strategies raises questions the authors’ results begin to answer but that future work could refine: how the threshold should be calibrated across different material families, how the approach behaves with very small target datasets, and whether the same similarity measure remains informative as feature representations evolve. The study was received in April 2025 and accepted in September 2026, and it is published open access under a Creative Commons Attribution 4.0 license, making the full methodological detail available to practitioners who wish to adapt the approach to their own property prediction problems.

Subject of Research: AMS-LPP: adaptive model selection for lightfastness property prediction in textile materials using similarity-guided transfer learning and AutoML

Article Title: AMS-LPP: adaptive model selection for lightfastness property prediction in textile materials using similarity-guided transfer learning and AutoML

Article References: Khan, M. A., Faseeh, M., Naqvi, S. S. A., & Kim, D.-H. (2026). AMS-LPP: adaptive model selection for lightfastness property prediction in textile materials using similarity-guided transfer learning and AutoML. Journal of Big Data. https://doi.org/10.1186/s40537-026-01585-4

Image Credits: AI Generated

DOI: 10.1186/s40537-026-01585-4

Keywords: machine learning, materials science, textiles, transfer learning, AutoML, AMS-LPP, adaptive, model, selection, lightfastness, property, prediction

The core insight behind the work is that classification errors across classes can conflict. A network that excels at distinguishing shirts from coats in the Fashion MNIST dataset might struggle with sneakers versus sandals, and the conventional practice of averaging errors into a single scalar loss hides these tensions entirely. By contrast, the CMaO-SL formulation defines one objective per class, so a ten-class problem becomes a ten-objective optimization problem. Because no single architecture can simultaneously minimize every class-specific error, the search targets a Pareto set: a collection of non-dominated solutions in which improving performance on one class necessarily comes at some cost to another. This trade-off information, normally invisible to practitioners, becomes an explicit output of the training process.

Solving such a problem is computationally daunting. Not only must the search evaluate many candidate architectures, but it must do so repeatedly to draw statistically meaningful conclusions about a stochastic process. The authors tackled this challenge with a two-pronged strategy combining transfer learning with evolutionary neural architecture search. Every network in their framework is split into two parts: a deep, pre-trained feature extractor kept completely frozen, and a shallower classification sub-network that is allowed to evolve. The feature extractor is derived from a ResNet-18 network pre-trained on ImageNet, with its final layers removed and its pooling adjusted to output 2,304 features. Because the backbone is never retrained or fine-tuned, the expensive part of the computation is paid only once, and the search can concentrate on the smaller, cheaper classification head.

The evolutionary search itself adapts MOEA/D, a well-established decomposition-based multiobjective evolutionary algorithm, to the architecture search setting. The user first defines a master architecture for the classification sub-network, consisting of fully connected layers with batch normalization and ReLU activations. Each individual in the evolving population is a binary mask vector that dictates which connections in the master architecture are pruned; where a mask entry is zero, the corresponding connection is removed and its weight fixed at zero. This pruning-as-search strategy generates a rich family of alternative architectures from a single blueprint. Each candidate is then trained with gradient descent, evaluated per class on a validation set, and assigned a performance vector whose components are one minus the F1 score for each class.

The evaluation procedure embeds an early-stopping mechanism to guard against overfitting. During each candidate’s five training epochs, the algorithm tracks a scalar validation error and retains the best-performing weight vector, ensuring that later epochs cannot degrade the network that gets assessed. The authors used two training loss functions in their experiments, cross-entropy and a weighted average of per-class binary cross-entropy aligned with each sub-problem’s reference direction, and two corresponding validation error functions. This produced five distinct search configurations, three evolutionary and two non-evolutionary, the latter serving as controls in which independent scalar searches replaced the coupled evolutionary process.

The experimental study spanned six benchmark datasets drawn from very different domains: Fashion MNIST, K-MNIST, CIFAR10, and three medical image collections from the MedMNIST suite, namely BloodMNIST, DermaMNIST and OCTMNIST. Each configuration was run fifteen times, allowing a rigorous statistical treatment that is rare in the Pareto-based architecture search literature. After confirming with a Shapiro-Wilks test that the results were not normally distributed, the authors compared configurations using medians and pairwise Mann-Whitney-Wilcoxon tests at a significance level of 0.05, alongside the IGD+ performance indicator, which measures how well an obtained front approximates a reference Pareto front.

The results were consistent across datasets and evaluation sets. Networks trained with cross-entropy loss produced better-approximated Pareto fronts than those trained with the weighted binary cross-entropy variant, and the evolutionary searches matched or beat their non-evolutionary counterparts on front quality while producing sets of networks with lower total complexity. Parallel coordinate plots of per-class F1 scores made the conflicting nature of the objectives vividly visible: the line connecting a classifier’s best performance on one class typically dipped on others. Crucially, when the authors picked the single best network from each evolutionary run’s set, measured by overall validation accuracy, those networks matched the accuracy of the unpruned master architecture while being substantially smaller, demonstrating that the many-objective formulation finds leaner models without sacrificing performance.

The benchmark comparisons are perhaps the most striking part of the study. When the networks discovered by the proposed method were plotted against published reference results for ResNet-18-based classifiers on the same datasets, no published result dominated the new networks on the combined axes of accuracy and complexity. The full ResNet-18 carries roughly 11.5 million parameters relevant to the comparison, whereas the evolved networks came in at around 2.5 million, meaning the reference architecture is about 360 percent more complex. That dramatic reduction in size cost less than 8 percent in accuracy, and on the DermaMNIST skin-lesion dataset the evolved classifiers actually surpassed the published benchmark. For a method that never fine-tuned its feature extractor and trained each candidate for only five epochs, this is a remarkable showing.

The practical implications extend beyond raw benchmark numbers. Because the output of the search is a set of classifiers with exposed trade-offs, users gain flexibility that single-model pipelines cannot offer. A deployment with tight computational budgets can select the smallest acceptable network from the set, while an application where a particular class matters most, say a rare disease category in a medical screening tool, can pick the member that favors that class. Alternatively, the trade-off structure can guide the construction of ensembles, an avenue the authors identify as a priority for future work. They also flag several open questions: whether fine-tuning the shared feature extractor would push accuracy higher, how the approach handles imbalanced data and noisy labels common in real-world problems, and how alternative pruning schemes, such as removing whole layers or blocks, might perform.

The study also fills a conceptual gap in the literature. Although Pareto-based multiobjective machine learning has existed for years, formal treatments of many-objective supervised learning, where the number of objectives exceeds three, have been scarce, and no prior work had combined such a formulation with neural architecture search and a transferred feature extractor. Most existing multiobjective NAS studies juggle only two or three objectives, typically accuracy, model size and inference cost, rather than one objective per class. By demonstrating a working pipeline at ten objectives, backed by fifteen-run statistical analysis per configuration, the Tel-Aviv and Braude team has shown that class-aware, many-objective architecture search is not only feasible but competitive. As automated machine learning matures, approaches like this suggest a future in which the question is no longer simply how accurate a network is, but which compromises it makes, and whether those are the compromises its users actually want.

Subject of Research: Class-based many-objective neural architecture search for deep classifiers using evolutionary optimization and transfer learning

Article Title: Class-based many-objective neural architecture search for deep classifiers

Article References: Mardiks, G., Salih, A., & Moshaiov, A. (2026). Class-based many-objective neural architecture search for deep classifiers. Neural Computing and Applications, 38(19), Article 800. https://doi.org/10.1007/s00521-026-12510-4

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12510-4

Keywords: neural architecture search, many-objective optimization, Pareto front, evolutionary algorithms, transfer learning, deep learning, multi-class classification, MOEA/D, pruning, ResNet-18, AutoML, machine learning

The research team, led by Ashiqur Rahman Khan and colleagues, built their predictive models using the Nigeria Malaria Indicator Survey 2021, a nationally representative dataset that captures demographic, household, and socioeconomic characteristics across the country. This choice of data source is significant. In many resource-limited settings, rapid diagnostic tests are the frontline screening tool, but they are not perfect. False-negative results can send an infected child home without treatment, and delays in confirmatory microscopy can postpone care until the disease has progressed. A statistical model that flags high-risk children using information already collected in routine surveys could serve as an early warning layer that complements, rather than replaces, conventional diagnostics.

The methodological pipeline began with 43 candidate variables drawn from the survey. Through careful preprocessing and statistical dependency testing, the team whittled this list down to 13 statistically significant predictors. This dimensionality reduction is more than a housekeeping step; it guards against overfitting, the phenomenon in which a model memorizes noise in the training data rather than learning genuine patterns that generalize to new populations. By retaining only variables with demonstrated statistical relevance, the researchers improved the odds that their models would perform reliably on unseen data.

Five machine learning approaches were then trained and compared: Logistic Regression, a Decision Tree, a Support Vector Classifier, Extreme Gradient Boosting, and H2O AutoML, a platform that automatically searches across many model configurations to find strong candidates. Evaluation followed rigorous practice, using stratified train-test splitting to preserve class proportions and 10-fold cross-validation to ensure that performance estimates were not artifacts of a single lucky split. Because missing a true malaria case is far more dangerous than a false alarm, the researchers emphasized the F2-score, a metric that weights recall, the ability to catch actual positives, more heavily than precision.

On the full dataset, the H2O AutoML Generalized Linear Model achieved the highest F2-score at 79.38 percent, with a recall of 83.08 percent, meaning it correctly identified more than four out of five children who truly belonged in the high-risk category. XGBoost, the gradient-boosted ensemble famous for dominating tabular data competitions, posted the highest test accuracy at 75.54 percent. The contrast between these two results illustrates a classic tension in medical machine learning: overall accuracy treats false positives and false negatives as equally costly, while a recall-oriented metric reflects the clinical reality that a missed infection can be fatal.

One of the study’s most interesting innovations addresses a subtle but important problem: regional confounding. Nigeria’s malaria epidemiology varies dramatically by geography, driven by differences in rainfall, ecology, and intervention coverage. A model that simply learns regional identifiers risks encoding where a child lives rather than the underlying risk factors, producing a tool that may not transfer to new regions or that obscures the true drivers of risk. To mitigate this, the team performed Agglomerative Hierarchical Clustering on the eight most important predictors after deliberately excluding regional identifiers, allowing the data to reveal natural population subgroups on its own.

The clustering analysis yielded two optimal population subgroups, with a Silhouette Score of 0.3939, a moderate but acceptable indication that the two clusters are internally coherent and distinct from one another. Rather than treating the entire nation as a homogeneous population, the researchers then trained cluster-specific models. The results validated the strategy: in Cluster 1, the H2O Generalized Linear Model delivered the best recall-oriented performance with an F2-score of 83.31 percent, while in Cluster 2, XGBoost took the lead with an F2-score of 83.25 percent. Both cluster-specific figures exceed the best overall-dataset F2-score, demonstrating that accounting for population heterogeneity measurably improves predictive performance.

The implications for public health practice in Nigeria and beyond are considerable. Malaria control programs already collect rich household survey data through instruments like the Malaria Indicator Survey, yet these data are often analyzed only in aggregate for policy planning. This study shows that the same data can be repurposed into an operational risk-stratification tool. Health workers equipped with a model trained on such data could prioritize households for bed net distribution, targeted testing campaigns, or preemptive treatment, stretching scarce resources further in settings where every diagnostic kit and every clinic visit counts.

The choice of H2O AutoML as the top performer also carries practical weight. Automated machine learning platforms reduce the expertise barrier, allowing public health analysts without deep data science backgrounds to retrain models as new survey waves arrive. In a country where malaria transmission dynamics shift with seasons and intervention campaigns, a framework that can be refreshed regularly, without a team of specialists hand-tuning algorithms, is far more likely to be sustained in practice than a bespoke model that decays as conditions change.

Caveats remain, as they do with any model built on survey data rather than clinical measurements. The predictors capture demographic, household, and socioeconomic conditions, not parasitological evidence, so the framework identifies risk rather than confirming infection, and it is designed to complement rapid diagnostic tests and microscopy, not to replace them. Still, the study demonstrates a robust and automated pathway toward early malaria risk assessment in the populations that need it most. If validated prospectively in field settings, cluster-aware AutoML screening could become a quiet but powerful addition to the toolkit fighting one of the world’s oldest and deadliest diseases.

Subject of Research: Machine learning-based early malaria risk prediction in Nigerian minors using non-clinical survey data

Article Title: Early malaria risk screening in Nigerian minors using AutoML and cluster-based analysis of non-clinical survey data

Article References: Khan, A. R., Mahbub, N., Aziz, R. A., Siddique, M. H., & Sayeem, Z. I. (2026). Early malaria risk screening in Nigerian minors using AutoML and cluster-based analysis of non-clinical survey data. PLOS Digital Health, 5(9), e0001736. https://doi.org/10.1371/journal.pdig.0001736

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001736

Keywords: malaria, Nigeria, AutoML, machine learning, risk prediction, hierarchical clustering, XGBoost, public health, Malaria Indicator Survey, rapid diagnostic tests, children under five, resource-limited settings

The research, published in the Journal of New Approaches in Educational Research, tackles a problem that has long frustrated educational data scientists: with so many machine learning models available, each with its own tuning requirements, how does an institution without a dedicated data science team find the one that works best for its data? The authors’ answer was to let an algorithm do the searching. Using the Auto-Weka framework, which automates both model selection and hyper-parameter optimization, the team allowed a search procedure to iterate through a long list of predictive strategies and their associated settings until it converged on the configuration that delivered the highest classification accuracy.

The search landed on an ensemble model, a strategy that combines the outputs of several base classifiers rather than relying on any single one. The ensemble evaluated in the study drew on five constituent techniques: artificial neural networks, K-nearest neighbors, Naive Bayes, support vector machines, and logistic regression. Each of these brings a distinct mathematical personality to the task. Neural networks learn layered, nonlinear representations of the data through weighted connections between artificial neurons. K-nearest neighbors classifies a new student by looking at the most similar labeled cases in the training set, using distance functions such as the Euclidean metric. Naive Bayes applies Bayesian probability under the simplifying assumption that features are conditionally independent, which makes it fast to train. Support vector machines construct an optimal separating hyperplane between classes, using kernel functions to handle data that cannot be split linearly. Logistic regression maps a linear combination of inputs through a sigmoid function to produce a probability between zero and one.

In the neural network component used here, the architecture consisted of an input layer representing the categorized data features, two hidden layers containing twelve and seven neurons respectively, and a single output neuron representing the binary outcome. The sigmoid activation function was chosen because it modulates values smoothly between zero and one, making it well suited to probability-style outputs. For the support vector machine, the researchers employed a polynomial kernel of a specified degree, which implicitly casts the data into a higher-dimensional space where, according to Cover’s theorem, a hyperplane is more likely to separate the two classes cleanly. These technical choices, normally the province of experienced practitioners, were arrived at automatically through the AutoML search rather than by manual trial and error.

The dataset behind the study came from the academic records of students enrolled in a university course at Damietta University between 2016 and 2021. The raw collection contained 480 records, but after preprocessing to remove outliers, missing values, and inconsistencies, 461 usable instances remained. The data was split into a training set of 329 instances, roughly seventy percent, and a test set of 132 instances, roughly thirty percent. Five features described each record: the academic year, the midterm score, the writing exam score, the final degree, and the overall grade. The grades were distributed across five categories from A to F, with D grades, representing a pass, the most common at nearly thirty-eight percent of students, followed by F grades, representing failure, at nearly twenty-nine percent.

The failure statistics buried in those records are striking. In 2016, 70.21 percent of students in the course failed, and in 2017 the rate climbed to a peak of 71.21 percent. The picture improved dramatically in later years, with failure rates of 19.66 percent in 2018, 9.78 percent in 2019, 16.13 percent in 2020, and 17.07 percent in 2021, but the early years illustrate exactly why institutional decision makers want early-warning tools. Only twelve students across the entire five-year span achieved the top A grade, a mere 2.6 percent of the sample, underscoring how demanding the course was and how much room there is for targeted intervention.

When the researchers benchmarked seven classification methods on the test set, the ensemble approaches dominated. Bagging, which trains multiple models on resampled subsets of the data and aggregates their votes, classified all 132 test instances correctly, achieving one hundred percent accuracy along with perfect precision, recall, F-measure, and kappa statistics. Random Forest, a related ensemble built from decision trees, misclassified just one instance, yielding 99.26 percent accuracy and scores of 0.99 across the other metrics. Naive Bayes followed at 95.68 percent accuracy, then the artificial neural network at 91.6 percent, logistic regression at 90.13 percent, K-nearest neighbors at 89.82 percent, and the support vector machine at 79.3 percent. The kappa coefficients, which correct for agreement that could occur by chance, told the same story, ranging from 0.74 for the support vector machine to a perfect 1.0 for Bagging.

The comparison with earlier literature is instructive. Previous studies had reported Random Forest accuracy of 72.4 percent and Naive Bayes accuracy of 88.3 percent on comparable tasks, while K-nearest neighbors had reached 92.6 percent in one survey. The substantially higher figures in the current work suggest that the combination of careful preprocessing, the AutoML-driven selection of hyper-parameters, and the intrinsic strength of ensemble methods can push performance well beyond what individual classifiers typically achieve. The authors attribute the ensemble advantage to the interdependencies among features: because the predictors are correlated, combining multiple base learners that each capture different aspects of those relationships produces a more robust and more accurate overall model than any single technique can manage alone.

The practical implications extend beyond the leaderboard. The researchers frame the predictive model as a decision-support tool for medical sector colleges, one that could inform amendments to admission systems and student selection methods using statistics and grades accumulated over the preceding five years. Identifying weak students early, particularly in the first year when dropout risk peaks, would allow institutions to intervene without lowering educational standards. The study is candid about its limitations, however: the dataset covered a single course at one institution, contained only five features, and relied on records from students who had already begun their studies rather than applicants. The authors note that most published work using methods like Naive Bayes similarly depends on data from enrolled students, which limits how early in the pipeline predictions can be made.

Future work, the team writes, will expand both the number of features and the number of instances in the dataset to enable deeper analysis of educational data, with the goal of better distinguishing struggling students from thriving ones and ultimately reducing failure rates. They also plan to layer optimization techniques such as differential evolution and genetic algorithms onto the predictive framework, potentially squeezing out further gains. For now, the study stands as a compelling demonstration that automated machine learning can compress what used to be a specialist’s weeks of model tuning into an algorithmic search, and that when it comes to forecasting the academic fate of medical students, the wisdom of many models combined decisively outperforms the judgment of any one.

Subject of Research: Automated machine learning for predicting academic performance of medical students

Article Title: Predicting student performance academic using Automated Machine Learning (AutoML): in medical academic institutions

Article References: Abougalala, R. A., Alharbi, N., Amasha, M. A., Areed, M. F., Alkhalaf, S., & Khairy, D. (2025). Predicting student performance academic using Automated Machine Learning (AutoML): in medical academic institutions. Journal of New Approaches in Educational Research, 14(1), Article 19. https://doi.org/10.1007/s44322-025-00038-9

Image Credits: AI Generated

DOI: 10.1007/s44322-025-00038-9

Keywords: AutoML, machine learning, student performance prediction, medical education, ensemble methods, Random Forest, Bagging, educational data mining, Damietta University, classification algorithms, dropout prediction, learning analytics

The research, conducted by Pabitha B and Vani V of the Department of Computer Science and Engineering at the National Institute of Technology Puducherry in Karaikal, India, together with Shridhar Sanshi of the National Institute of Technology Karnataka in Surathkal, began with a deceptively simple question: how do you know whether the biological data flowing through a body area network is healthy or compromised? The team’s answer involved building a wearable WBAN model around a consumer smartwatch, using it to estimate a suite of vital signs including heartbeat, pulse, oxygen intake and exhalation, blood pressure, and oxygen saturation. These measurements, gathered from real participants who gave informed consent, formed the raw material for a machine learning experiment with a twist.

The twist is automation. Conventional machine learning pipelines demand that a human expert choose an algorithm, tune its hyperparameters, preprocess the data in the right order, and iterate until performance plateaus. That process is slow, labor-intensive, and highly dependent on the skill of the practitioner. The Indian team instead turned to the Tree-based Pipeline Optimization Tool, or TPOT, an open-source AutoML framework that uses genetic programming to evolve entire machine learning pipelines. Rather than a single classifier, TPOT assembles and optimizes chains of preprocessing steps and models, treating each candidate pipeline as an organism in an evolutionary population. Over successive generations, the fittest pipelines survive, mutate, and recombine, until the tool converges on a configuration that performs best on the validation data.

Applied to the smartwatch-generated dataset, this evolutionary search produced a striking result: the best pipeline TPOT discovered was built around the XGBoost classifier, a gradient-boosted decision tree ensemble that has become a workhorse of tabular data science. When tested on held-out data, this evolved pipeline classified vital-sign records as normal or abnormal with 98.91 percent test accuracy. The researchers compared this automated approach against traditional, manually configured machine learning classifiers across a battery of standard metrics, including precision, accuracy, recall, and F1 score. The comparison revealed that TPOT not only matched or exceeded the conventional models in classification quality but also reduced the complexity and processing overhead of anomaly detection, a critical advantage when the goal is high-speed diagnosis in real-world clinical settings.

The technical significance of this result lies in what anomaly detection actually means in a WBAN context. Unlike a hospital monitor tethered to a wall, body area networks operate over wireless links that are inherently vulnerable. Malicious actors can inject false readings, replay old data, or degrade signal integrity, and even benign faults, such as a poorly seated sensor or a dropped packet, can produce values that look pathological. The team framed the problem as a binary classification task: given a record of vital signs, decide whether the WBAN data is healthy or unhealthy. By training on real-time data collected from the smartwatch deployment, the classifiers learned the statistical fingerprint of genuine physiological readings, and deviations from that fingerprint, whether caused by attack or artifact, could be flagged automatically before the data reaches a diagnostic decision point.

The choice of vital signs was deliberate. Heart rate and pulse provide complementary windows into cardiovascular function, while blood pressure and oxygen saturation are among the most clinically actionable parameters in acute care. Respiratory measures, captured as oxygen intake and exhalation, round out the picture. The authors grounded their labeling of normal versus abnormal values in established clinical guidelines, referencing published vital-sign parameter standards and widely used medical references for pulse-rate thresholds. This anchoring matters, because an anomaly detector is only as good as its definition of normal; without clinically validated boundaries, a model risks learning idiosyncrasies of its own dataset rather than medically meaningful deviations.

What makes the study particularly timely is the explosion of the Internet of Medical Things, often abbreviated IoMT, in which wearables, ingestible sensors, and connected devices generate continuous health data at scale. Prior work in this space has ranged from smart jackets that detect early hypothermia symptoms to ingestible sensors that verify medication adherence, and from convolutional neural networks that classify abnormal WBAN data to intrusion detection schemes built on game theory and support vector machines. The literature reviewed by the authors shows a field converging on a consensus: machine learning is the most viable path to protecting patient data integrity, but the field has lacked a systematic way to find the best model for each deployment. AutoML addresses precisely that gap, effectively democratizing access to state-of-the-art pipelines without requiring every hospital or research group to employ a team of ML engineers.

The evolutionary mechanics of TPOT deserve a closer look, because they explain both the accuracy gain and the efficiency claim. TPOT represents each candidate pipeline as a tree structure encoding data transformations, feature selection steps, and a final estimator. A fitness function, typically cross-validated performance on the training data, scores each tree. Selection pressure then drives the population toward high-performing configurations, while crossover and mutation operators introduce new combinations of preprocessing and modeling steps that a human might never try. In this study, the process converged on an XGBoost-based pipeline, suggesting that the structure of the vital-sign data, with its mix of continuous physiological variables and potentially nonlinear interactions between them, rewards the kind of ensemble reasoning that gradient boosting provides. Crucially, once the pipeline is discovered, it can be exported as reusable code and applied to future data, so the expensive search happens once, and the deployed classifier runs efficiently at the edge.

That last point carries real weight for deployment. A WBAN node is battery-powered and computationally constrained; heavyweight inference on the device itself drains resources, while shipping raw data to the cloud raises latency and privacy concerns. By demonstrating that an AutoML-derived pipeline reduces processing overhead relative to traditional machine learning approaches, the researchers point toward anomaly detection that can run close to the source of the data, enabling the high-speed diagnosis their abstract emphasizes. The pipeline TPOT generates is not a black box in the architectural sense: it is an explicit, inspectable sequence of operations, which matters in healthcare, where clinicians and regulators increasingly demand transparency about how automated decisions are made.

The study is not without boundaries, and the authors are transparent about them. The dataset, collected from smartwatch devices worn by study participants, contains personal health information and is therefore not publicly available, in accordance with ethical guidelines protecting participant privacy; researchers seeking access may contact the corresponding author. The binary framing of healthy versus unhealthy data, while powerful, is a simplification of the richer taxonomy of faults and attacks a real network might face, and the 98.91 percent figure, though impressive, reflects performance on this specific deployment and dataset. Still, the work marks a meaningful step: it demonstrates that real-time, smartwatch-generated vital signs can be protected by an automatically discovered classifier, that the resulting pipeline generalizes to future data, and that the approach trims the computational cost that has historically made rigorous anomaly detection impractical at the network’s edge.

The broader implication is one that patients, clinicians, and device manufacturers should all take note of. As wearables evolve from fitness accessories into genuine medical instruments, the integrity of the data they produce becomes a patient-safety issue, not merely a data-quality nuisance. Automated machine learning offers a way to keep pace with that shift, continuously discovering and refreshing the models that stand guard over our most intimate measurements. If the body area network is the nervous system of remote medicine, then tools like TPOT are beginning to serve as its immune system, evolving in response to threats so that the signals reaching a doctor’s screen can be trusted, and the diagnosis built upon them can be made with confidence.

Subject of Research: Automated machine learning-based anomaly detection of vital-sign data in wireless body area networks

Article Title: Automated Anomaly Detection in WBAN Smartwatch-Generated Vital Signs Using TPOT Classifier

Article References: B, P., V, V., & Sanshi, S. (2026). Automated Anomaly Detection in WBAN Smartwatch-Generated Vital Signs Using TPOT Classifier. Mobile Networks and Applications. https://doi.org/10.1007/s11036-026-02525-5

Image Credits: AI Generated

DOI: 10.1007/s11036-026-02525-5

Keywords: WBAN, AutoML, TPOT, anomaly detection, smartwatch, vital signs, machine learning, XGBoost, IoMT, wearable technology, healthcare security, data integrity

The review, authored by Xinyu Li, Zhaofa Zhou, Zhili Zhang, Zhenjun Chang, Zheng Lian, Hui Duan and colleagues at Rocket Force Engineering University in Xi’an, China, positions TFNAS as a distinct and emerging research paradigm within automated machine learning. Neural architecture search has long served as what the authors call a performance engine for AutoML, driving intelligent advancements across domains from computer vision to speech processing. Yet its development has been significantly constrained by prohibitively high computational costs, a bottleneck that has kept sophisticated architecture discovery largely out of reach for all but the best-funded laboratories.

The core insight behind TFNAS is deceptively simple. Rather than training each candidate network to convergence and measuring its accuracy, the method scores architectures at initialization or after only minimal computation, using proxy metrics that correlate with final performance. Metrics in this family include analyses of network expressivity, progress indicators measured without gradient updates, and zero-cost probes that estimate how well a freshly initialized model will learn. Because no candidate network is trained inside the search loop, thousands of architectures can be screened in the time it would take to train just one.

Importantly, the review offers a careful definition of what training-free actually means. In its strict sense, the term refers to the absence of candidate network training during the search process itself. In a broader interpretation, it also permits the offline pre-training of lightweight performance predictors, so long as those predictors do not constitute training at the search stage. This distinction matters, the authors argue, because it clarifies the boundary of the field and prevents confusion with related approaches such as one-shot and supernet-based methods, which still rely heavily on weight sharing and partial training.

To bring order to a rapidly proliferating literature, the review proposes a two-dimensional taxonomy organized around the source of information used by a proxy metric and the methodology by which that metric is constructed. The information-source dimension captures whether a metric draws on architectural properties, weight statistics, activation patterns, gradient signals or learned predictors. The construction dimension distinguishes metrics that are hand-designed from theoretical principles, those derived empirically from large collections of trained architectures, and hybrid approaches that combine both. The authors use this framework to trace the developmental trajectory of state-of-the-art techniques, showing how the field has moved from simple initialization statistics toward increasingly sophisticated, multi-signal evaluation schemes.

One of the review’s most consequential claims concerns the co-evolution of evaluation metrics and search strategies. The authors contend that the two components should not be designed in isolation. Instead, they posit that joint optimization characterized by dynamic feedback and mutual reinforcement is key to maximizing efficacy. In practice, this means a search algorithm can be tuned to exploit the strengths and tolerate the weaknesses of its scoring function, while the scoring function can be refined in light of where the search actually explores. This evolutionary mechanism, the review suggests, explains why some TFNAS pipelines dramatically outperform others even when built from superficially similar parts.

The promise is substantial. If architecture evaluation that once required days of GPU time can be compressed into seconds, automated machine learning becomes accessible to researchers without massive compute budgets, and the design cycle for deployment-critical models on edge devices, smartphones and embedded systems shrinks dramatically. The technique also reduces the energy footprint of architecture discovery, an increasingly pressing concern as the computational appetite of AI research continues to grow. By lowering the barrier to entry, TFNAS could democratize a layer of AI engineering that has been effectively reserved for industrial giants.

Yet the review is candid about the obstacles. The authors identify four core contradictions that persist as the technology moves toward practical application. These include the persistent gap between what cheap proxy metrics measure and what real trained performance requires, the risk that a metric optimized for one architecture family or dataset fails to generalize elsewhere, tensions between evaluation speed and evaluation fidelity, and the difficulty of integrating training-free scores into search strategies that were originally conceived around trained validation accuracy. While the article’s abstract names these contradictions without exhaustively detailing each, the authors make clear that resolving them is the central task standing between current research and reliable, widely applicable deployment.

The review also outlines future directions for the field. These include developing more robust and interpretable proxy metrics, designing search strategies purpose-built for training-free signals, and establishing standardized benchmarks that would allow fair comparison across the fragmented landscape of published methods. The emphasis on interpretability echoes a broader concern in machine learning: practitioners are more likely to trust a zero-cost score if they understand why it predicts performance, rather than treating it as an opaque oracle. The authors frame their taxonomy and analysis as a foundation intended to guide that maturation process.

Published open access on 11 September 2026 after being received in March and accepted in early September, the article arrives at a moment when the AI community is scrutinizing the cost of progress. By systematically mapping what training-free architecture search has achieved, where it falls short, and how its evaluation and search components can be co-designed, the review offers both a reference work and a research agenda. If its diagnosis is right, the next generation of neural networks may be discovered not by brute-force training marathons, but by fast, principled measurement, a shift that could reshape who gets to build AI and how quickly the technology evolves.

Subject of Research: Training-free neural architecture search methods for reducing the computational cost of automated machine learning

Article Title: Systematic review of training-free neural architecture search: research progress, core challenges, and future directions

Article References: Li, X., Zhou, Z., Zhang, Z., Chang, Z., Lian, Z., & Duan, H. (2026). Systematic review of training-free neural architecture search: research progress, core challenges, and future directions. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11701-8

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11701-8

Keywords: neural architecture search, training-free evaluation, automated machine learning, proxy metrics, NAS, deep learning, AutoML, computational cost, search strategy, interpretability, model design, practical deployment