The research, conducted by Ghazi Bahroz Jumaa of the Civil Engineering Department at the University of Garmian in the Kurdistan Region of Iraq, tackles one of the most safety-critical questions in reinforced concrete design: how much shear force can a beam resist before it fails catastrophically? Shear failure is notoriously brittle and sudden, unlike the gradual, warning-giving flexural failure that designers prefer to govern. For beams without stirrups, the closed steel hoops that normally provide shear reinforcement, the entire resistance must come from the concrete itself, making accurate prediction of that resistance a matter of structural life and death.

The heart of the problem lies in the messy, heterogeneous nature of RCA. When old concrete is crushed, each recycled particle is a composite: natural stone wrapped in a layer of adhered mortar that is often weaker, more porous, and riddled with micro-cracks from the crushing process. In a loaded beam, diagonal cracks propagate through the concrete web, and the rough crack faces resist sliding through a mechanism engineers call aggregate interlock. Weaker, more irregular recycled particles interlock less effectively, and the old mortar attached to them introduces additional weak zones. The result is a material whose shear behavior is inherently more variable than that of natural-aggregate concrete, and whose performance depends on the percentage of recycled content, a variable that most established design equations were never formulated to include.

Existing design provisions, such as those in the American Concrete Institute’s ACI 318 code and Europe’s Eurocode 2, were calibrated decades ago on natural-aggregate concrete. When applied to RCA beams, they can be either unconservative or wastefully conservative, and the experimental literature over the past two decades, from studies by Han and colleagues in 2001 through work by Rahal and Alrefaei, Ignjatović and co-workers, and many others, has documented a wide scatter in measured shear strengths. Empirical equations such as the classic Zsutty formulation, dating back to 1971, capture broad trends but struggle with the added complexity of recycled content. This is precisely the kind of nonlinear, multi-variable prediction problem where machine learning has begun to shine.

Jumaa assembled a database of 179 experimental RCA beam specimens drawn from published laboratory studies spanning more than two decades. Each specimen was characterized by the key parameters that govern shear capacity: beam width, effective depth, the shear span-to-depth ratio, the longitudinal reinforcement ratio, the concrete compressive strength, and, crucially, the percentage of recycled concrete aggregate replacing natural stone. The measured shear capacities ranged from small laboratory beams failing at around 12 kilonewtons to massive 400-millimetre-wide specimens carrying more than 860 kilonewtons, giving the models a rich and demanding training landscape.

Five machine learning algorithms were trained and compared under a rigorous validation protocol: k-Nearest Neighbors, Random Forest, Support Vector Machines, Gaussian Process Regression, and Artificial Neural Networks. Rather than relying on a single train-test split, which can flatter a model by luck of the draw, the study employed a 10-fold cross-validation scheme repeated 20 times. In this procedure the data are repeatedly shuffled and divided into ten subsets; the model trains on nine and is tested on the held-out tenth, with the process cycled through all folds and then repeated from scratch twenty times to average out randomness. This is the gold standard for demonstrating that a model’s accuracy reflects genuine learning of underlying physics rather than memorization of individual specimens.

The results were striking. Support Vector Machines emerged as the top performer, achieving a coefficient of determination of 0.933, meaning the model explained more than 93 percent of the variance in measured shear strength, with a root mean square error of 29.98 kilonewtons and a mean absolute error of just 13.08 kilonewtons. Gaussian Process Regression followed closely with an R-squared of 0.912 and the lowest mean absolute error of the trio at 11.81 kilonewtons, while Artificial Neural Networks achieved an R-squared of 0.892. All three outperformed both the newly proposed empirical equations, including a modified Zsutty equation and an ACI 318-style equation incorporating RCA content, and the traditional design codes themselves. Support Vector Machines, which work by mapping inputs into a high-dimensional feature space where a maximum-margin boundary separates the data, proved especially adept at handling the nonlinear interactions between beam geometry, material strength, and recycled content.

Beyond raw accuracy, the study included a parametric analysis that interrogated how the trained models respond when individual variables change, a crucial sanity check for any data-driven engineering tool. The machine learning predictions reproduced physically sensible trends: shear capacity increased with beam depth and width, and decreased as the recycled aggregate content rose, consistent with the known degradation of aggregate interlock in RCA concrete. This alignment between statistical learning and structural mechanics matters enormously, because a black-box model that contradicts physical intuition would be dangerous to trust, no matter how impressive its error metrics appear on paper.

The practical implications reach well beyond the laboratory. As governments worldwide push circular-economy mandates and green building standards, the volume of RCA entering structural applications is set to grow sharply, and the fib Model Code 2020 and the next generation of Eurocode 2 are actively working toward codified design rules for recycled aggregate concrete structures. Reliable predictive tools are the prerequisite for such codification: engineers need confidence that a beam designed with, say, 50 or 100 percent recycled coarse aggregate will carry its intended shear load with the same margin of safety as a conventional beam. A validated machine learning model, trained on nearly two hundred real experiments, offers exactly that confidence, and could inform both code committees drafting new provisions and practitioners evaluating existing structures built with sustainable materials.

The study also fits into a broader movement sweeping through structural engineering, in which data-driven methods complement, and sometimes surpass, hand-derived empirical formulas. Researchers have already applied neural networks and related techniques to predict the shear capacity of fiber-reinforced polymer concrete beams, the compressive strength of lightweight and recycled concretes, and bond strength in reinforced concrete members. What distinguishes the present work is its focus on the specific, under-addressed case of stirrup-free RCA beams and its head-to-head comparison of five algorithms under a demanding cross-validation protocol. As Jumaa notes in the paper, the work represents a step forward in using machine learning to enhance structural engineering, with the potential to make a significant impact in the field of sustainable materials. If the construction industry is to recycle its own waste at scale, it may well need algorithms, as much as new mix designs, to make the resulting structures safe.

Subject of Research: Machine learning prediction of shear strength in reinforced recycled aggregate concrete beams without stirrups

Article Title: Shear strength prediction of reinforced recycled aggregate concrete beams without stirrups using soft computing and empirical methods

Article References: Jumaa, G. B. (2026). Shear strength prediction of reinforced recycled aggregate concrete beams without stirrups using soft computing and empirical methods. Neural Computing and Applications, 38(19), Article 771. https://doi.org/10.1007/s00521-026-12488-z

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12488-z

Keywords: machine learning, recycled concrete aggregate, shear strength, reinforced concrete beams, stirrups, Support Vector Machines, artificial neural networks, Gaussian Process Regression, ACI 318, Eurocode 2, sustainable construction, structural engineering

The stakes are far from academic. Industrial anomaly detection is a cornerstone of modern manufacturing, where catching the faint acoustic signature of a worn bearing or a misaligned belt before it fails can save enormous maintenance costs and prevent catastrophic downtime. Acoustic and vibration monitoring is especially attractive because it is non-intrusive and sensitive to subtle mechanical degradation. Yet classical pipelines built on support vector machines hit a wall when multiple anomaly types coexist and their feature distributions intermingle in Euclidean space. The result is an unacceptably high false positive rate, which in practice means operators learn to ignore the alarms, defeating the entire purpose of the monitoring system.

The researchers built their study around two physical test rigs with deliberately contrasting difficulty. The first, an Open Belt Drive (OBD) system, consists of rubber and metal chain belt drives in which wooden chopsticks are inserted through pre-drilled holes, producing sudden, loud breaking sounds that define the anomaly class. The second, a Mini 4WD (M4W) course, sends a miniature four-wheel drive vehicle around a three-lane track where it first strikes a one-millimeter wooden stick, generating a small impact sound, and then passes over Magic Tape, producing a faint scratching noise. While chopstick breaks are obvious even to human ears, the stick and tape sounds of the M4W rig are extremely difficult to discern, and the two anomaly types occur sequentially within each roughly five-second lap.

From five-minute recordings of each system, the team extracted 30 normal samples per dataset and converted the audio into compact feature vectors using autoregressive (AR) modeling. Each 10-second segment was resampled from 48 kHz to 1100 Hz, cut into non-overlapping 0.4-second windows, and characterized by AR coefficients estimated with the Yule-Walker method, with median aggregation across windows providing robustness against transient noise spikes. The AR model order was varied from 2 to 10, and information-theoretic criteria, the Akaike Information Criterion and Bayesian Information Criterion, independently identified orders 9 to 10 as optimal, validating the experimental range. Because real industrial equipment rarely fails often enough to supply labeled anomaly data, the team adopted an unsupervised One-Class SVM that learns a decision boundary from normal operation data alone, flagging significant deviations as anomalies.

Against this pipeline, three kernels competed: the classical RBF baseline, and two quantum kernel architectures. Quantum Kernel 1 (QK1) uses linear entanglement, with nearest-neighbor CNOT gates connecting adjacent qubits in a chain, creating pairwise correlations between consecutive features at a circuit depth that scales linearly with qubit count. Quantum Kernel 2 (QK2) employs all-to-all entanglement, connecting every qubit pair to capture global, higher-order dependencies at the cost of quadratic circuit depth. Both use a data re-uploading strategy in which AR coefficients are repeatedly encoded into rotation gates across two layers, producing quantum amplitudes that depend non-linearly on the input through interference. Importantly, the quantum kernels were implemented via classical simulation, using adjusted RBF gamma parameters that approximate the correlation structures of the two entanglement topologies, a choice the authors candidly flag as a limitation that leaves confirmation on actual quantum hardware to future work.

On the OBD dataset, where anomalies are cleanly separable, all three methods eventually achieved perfect classification, but the quantum kernels got there faster. QK1 reached a perfect F1 score of 1.0 with just four AR features, QK2 matched it at seven, while the RBF kernel needed eight. Wilcoxon signed-rank tests across the full feature range confirmed these differences were statistically significant, with p-values of 0.0420 for QK1 versus RBF and 0.0203 for QK2 versus RBF. In practical terms, the quantum kernels offered a computational efficiency advantage, roughly halving the feature extraction overhead, but no fundamental capability gap, since classical methods converged to the same asymptotic solution.

The M4W dataset told an entirely different story. There, QK2 achieved an F1 score of 0.893 with a false positive rate of just 0.033, while the RBF kernel collapsed to an F1 of 0.428 with a staggering false positive rate of 0.600, meaning 60 percent of normal operation was flagged as anomalous. The statistical significance was emphatic, with a p-value of 0.0023. QK1 landed in between at F1 = 0.661 but became unstable at higher feature orders. Most tellingly, the RBF kernel’s F1 trajectory remained flat, never exceeding 0.47 despite a fivefold increase in feature count, a signature of structural failure rather than insufficient data. ROC trajectory analysis made the mechanism visible: RBF wandered horizontally in ROC space, oscillating between false positive rates of 0.6 and 1.0 without ever approaching the ideal operating point, while QK2 systematically drove its false positive rate down from 0.93 to 0.03 as features were added.

To explain why, the researchers turned to t-SNE visualization of the AR feature space, backed by rigorous statistical testing. For OBD, Kruskal-Wallis tests confirmed complete class separation in both projected dimensions, with all pairwise Mann-Whitney comparisons passing Bonferroni-corrected significance thresholds. For M4W, the picture was one of fundamental inseparability: the second dimension showed no significant structure at all (H = 0.4, p = 0.942), and Magic Tape anomalies proved statistically indistinguishable from every other class. A striking paradox emerged in the decision score distributions: QK2 achieved its superior performance despite a centroid separation between normal and anomalous classes roughly three times smaller than RBF’s. The quantum kernel’s advantage lies not in pushing class means apart but in reshaping the distributions to minimize their overlap, exploiting non-linear correlations through high-dimensional Hilbert space projections that Euclidean geometry simply cannot represent.

The study also connects its unsupervised findings to earlier supervised work by the same group, including apple defect detection in shipping inspection that achieved F1 scores of 0.8 to 0.9 with only 24 training samples per class, far outperforming classical RBF. Across both paradigms, a consistent pattern emerges: quantum kernel benefits appear in small-data regimes where classical feature distributions overlap, while showing diminishing returns beyond moderate feature dimensionality, a phenomenon attributed to exponential concentration that may impose fundamental scalability limits. The authors propose a practical screening framework: run a preliminary t-SNE analysis with classical features, and if Kruskal-Wallis tests reveal poor separation or multiple inseparable class pairs, quantum kernels may be worth the investment. With false alarms triggering unnecessary inspections and production stoppages, a drop from a 60 percent to a 3 percent false positive rate is not a marginal improvement but the difference between a monitoring system operators trust and one they switch off. As quantum hardware matures, this work suggests that messy, overlapping, real-world sensor data, not pristine synthetic benchmarks, may be where quantum machine learning first earns its keep.

Subject of Research: Quantum kernel methods for unsupervised industrial acoustic anomaly detection with overlapping data distributions

Article Title: Potential of multi-anomalies detection using quantum machine learning

Article References: Tomono, T., & Tsujimura, K. (2026). Potential of multi-anomalies detection using quantum machine learning. Quantum Machine Intelligence, 8(2), Article 106. https://doi.org/10.1007/s42484-026-00448-8

Image Credits: AI Generated

DOI: 10.1007/s42484-026-00448-8

Keywords: quantum machine learning, quantum kernels, anomaly detection, One-Class SVM, acoustic monitoring, false positive rate, t-SNE, feature space, industrial condition monitoring, RBF kernel, entanglement, Hilbert space

The study, published in Smart Agricultural Technology, was led by Dthenifer Cordeiro Santana and colleagues at the Federal University of Mato Grosso do Sul (UFMS). The team ran two field experiments during the 2022/2023 growing season in the state of Mato Grosso do Sul, one at Nova França Farm in Costa Rica to produce leaves infected with target spot, and another at Gávea Farm in Chapadão do Sul to obtain rust-infected material. Both sites sit in a humid tropical climate on clay-textured Oxisol soils, and both were sown on October 15, 2022 with the cultivar M 5947 IPRO at a density of fourteen plants per meter. Rather than relying on natural infection alone, the researchers engineered a gradient of disease pressure through carefully timed fungicide programs—seventeen treatments for target spot and staged regimes for rust—so that leaves could be harvested at zero, twenty-five, and fifty percent symptomatic area.

Leaf collection took place at the R5.5 phenological stage, during grain filling, with one hundred leaves sampled for each severity level of each disease. The samples were classified against a standard diagrammatic severity scale, stored in polystyrene boxes to preserve turgor, and rushed to the UFMS Spectroscopy Laboratory in Chapadão do Sul. This ex situ approach was deliberate: by measuring leaves in the laboratory, the team could standardize illumination, acquisition geometry, and the sampled area while stripping out the confounding noise of canopy structure, soil background, shading, and field heterogeneity. It is the spectral equivalent of listening to a single violin in a soundproof room rather than in the middle of an orchestra.

Each leaf was scanned with a FieldSpec 3 Jr spectroradiometer using an ASD Plant Probe with its own light source, capturing reflectance from 350 to 2500 nanometers—three readings per leaf, averaged. That full spectrum yields 1025 individual data points per sample, a richness that is both a blessing and a computational burden. To tame it, the researchers tested two dimensionality-reduction schemes drawn from earlier work: the spectrum partitioned into 28 broader spectral bands, and a set of 22 reflectance inflection differences, or RIDs, which capture where the spectral curve changes slope. Principal component analysis on these reduced datasets explained eighty-five percent of total variance and immediately revealed structure: target spot was associated with essentially every spectral band, while soybean rust at fifty percent severity clustered with bands B11 through B22, spanning 501 to 710 nanometers, and with band B27 at 1930 nanometers in the short-wave infrared. Healthy leaves, by contrast, linked most strongly to the first three bands in the deep violet edge of the visible range.

The spectral signatures themselves told a biologically coherent story. Target spot produced the highest reflectance of any class across the entire spectrum, a consequence of the way its necrotic lesions destroy leaf architecture. As lesions progress from healthy tissue through affected tissue to dead, concentric necrotic zones, they degrade chlorophyll, disrupt the spongy mesophyll, and collapse the internal scattering structures that normally dominate near-infrared reflectance. Tissue necrosis is known to raise infrared reflectance, and the loss of pigment pushes reflectance up in the visible bands as leaves yellow. Fungal spore pigmentation may add further signal in the 580 to 700 nanometer range. Rust behaves differently: its uredinia, the pathogen’s reproductive structures, form within the leaf and alter photosynthesis and pigment content in surrounding tissue, producing more pronounced differences in the visible and near-infrared regions but far weaker separation in the short-wave infrared, where rust curves overlapped with healthy leaves at several points—likely because the pathogen, once established inside the leaf, does not dramatically extract water from it, and the SWIR region is closely tied to leaf water content.

With signatures established, the team put ten machine learning algorithms to work on three different inputs: the full spectrum, the 28 spectral bands, and the 22 RIDs. The lineup included multilayer perceptron neural networks, REPTree and J48 decision trees, Random Forest, Random Tree, Logistic Regression, AdaBoost, Bagging, and Sequential Minimal Optimization (SMO), a support vector machine implementation. All models were run in WEKA with default hyperparameters and evaluated by stratified ten-fold cross-validation repeated ten times, scored on correctly classified instances, the Kappa statistic, and the F-measure. The results formed a clear pattern. For disease identification, Logistic Regression and SMO fed the full spectrum were the champions, reaching roughly ninety-four percent correctly classified instances, a Kappa of about 0.9, and an F-measure near 0.9. When inputs were reduced, the multilayer perceptron and Logistic Regression took the lead, hitting around ninety percent accuracy with either spectral bands or RIDs.

Severity classification proved harder, as expected, but the models still performed impressively. Using the full spectrum, SMO alone stood out, achieving eighty-seven percent correctly classified instances and a Kappa of 0.85. With reduced inputs, the multilayer perceptron and Logistic Regression again excelled, reaching about eighty-three percent accuracy with RIDs and roughly eighty-two percent with spectral bands, with Kappa values around 0.78 to 0.79. The practical implication is significant: where computing power or sensor capability is limited, farmers and agronomists can trade a few points of accuracy for dramatically faster processing by using compressed spectral information, without abandoning the diagnostic pipeline altogether.

The authors were careful to address an obvious confound: could the fungicides themselves, rather than the diseases, have shaped the spectral signatures? They note that hyperspectral detection of fungicide deposits has been demonstrated in grapevine, so the possibility cannot be categorically dismissed. However, evidence from soybean suggests the effect is minor at recommended field rates. Previous studies found no significant impact of triazole-, strobilurin-, and carboxamide-containing fungicides on gas exchange, chlorophyll fluorescence, or yield in disease-free plants, and up to three consecutive applications across fifteen fungicide programs did not significantly alter soybean canopy reflectance in the visible and near-infrared. Crucially, leaves in this study were assigned to classes by their realized symptomatic area on a diagrammatic scale, not by treatment identity. Still, the team acknowledges that because fungicide regime and disease severity were not fully orthogonal—particularly in the rust experiment—the design cannot statistically separate fungicide effects from disease effects, and the models should be read as identifying spectral patterns associated with observed disease type and severity under the tested field conditions.

What elevates the work beyond a proof of concept is its integrated comparison of processing and classification strategies. Rather than simply confirming that hyperspectral data can distinguish sick plants—a result now well established for grapevine trunk diseases, tomato bacterial leaf spot, strawberry gray mold, and soybean frogeye leaf spot—the study maps how dimensionality reduction and input selection change the performance of each algorithm. That map matters for anyone building a real diagnostic tool, because it says which combinations to deploy under which constraints. The authors are candid about the next hurdles: the measurements were made on detached leaves in a laboratory, and real-world scouting will have to contend with canopy architecture, natural lighting, soil background, and shading. They call for future work using in situ field measurements, validation across independent locations and seasons, and a factorial pathogen-by-fungicide design that includes fungicide-treated pathogen-free plants and untreated infected plants to fully disentangle the signals.

Even with those caveats, the trajectory is unmistakable. A sensor that reads a leaf’s chemistry through its reflectance, coupled with algorithms that classify disease and severity in seconds, could transform phytosanitary management from reactive guesswork into precision intervention—spotting infections early, targeting fungicide applications only where needed, and slowing the resistance crisis that has made target spot such a formidable adversary. Detecting the lowest severity levels remains the field’s holy grail, and the authors flag it as a priority for future research. But the core demonstration stands: the spectral fingerprint of a diseased soybean leaf is rich enough, and machine learning is now sharp enough, that the eye of a trained pathologist may soon have a tireless, nanometer-precise rival.

Subject of Research: Hyperspectral sensing and machine learning for identifying soybean diseases and infection severity levels

Article Title: Beyond visual inspection: Smart identification of soybean diseases and infection severities using hyperspectral sensor

Article References: Santana, D. C., Baio, F. H. R., Otone, J. D. D. Q., Martins, E. V., Teodoro, L. P. R., Aptoula, E., & Teodoro, P. E. (2026). Beyond visual inspection: Smart identification of soybean diseases and infection severities using hyperspectral sensor. Smart Agricultural Technology, 15, Article 102502. https://doi.org/10.1016/j.atech.2026.102502

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102502

Keywords: soybean, hyperspectral sensing, machine learning, soybean rust, target spot, plant disease detection, precision agriculture, remote sensing, spectral signatures, support vector machines, dimensionality reduction, Brazil

The central premise of the review is deceptively simple: when abnormal examples are rare, expensive, or impossible to label, a model should learn from what is normal and flag everything else. This inversion of the usual supervised learning recipe has profound consequences for how algorithms are designed and evaluated. In conventional classification, decision boundaries are shaped by examples of every class, and the learner can exploit contrasts between them. In the one-class setting, the model sees only a single class — typically healthy operation — and must instead characterize the shape, density, or dynamics of that class so thoroughly that anything falling outside its description is treated as suspect. The review’s authors organize the field’s answers to this challenge into six methodological families: distance-based, boundary-based, density-based, reconstruction-based, feature-representation, and contrastive-representation approaches. Each family embodies a different philosophical bet about what “normal” looks like in temporal data, and each carries distinct strengths and failure modes that practitioners need to understand before deployment.

Distance-based methods are perhaps the most intuitive of the six. They rest on the assumption that normal observations cluster together in some appropriate space, so that an incoming time series segment can be judged by how far it lies from known normal examples or from prototypes summarizing them. Dynamic time warping, abbreviated DTW in the review’s extensive abbreviation list, plays a starring role here because it provides a way to compare sequences of different lengths and speeds — a necessity when heartbeats or machine cycles do not arrive on a rigid schedule. The appeal of these methods lies in their transparency: a practitioner can often inspect which normal example a new observation most resembles, or how far it deviates, and communicate that reasoning to domain experts. The cost is computational, since naive distance computations against large libraries of normal sequences scale poorly, and the choice of distance metric can quietly determine whether subtle temporal anomalies are visible at all.

Boundary-based methods take a different bet: rather than measuring distances to individual examples, they draw a closed envelope around the entire body of normal data. The one-class support vector machine, or OCSVM, is the canonical representative, mapping inputs into a high-dimensional feature space where a hyperplane or hypersphere can separate the normal region from the rest. The deep support vector data description, Deep SVDD, extends this idea by learning the mapping itself with a neural network, so that the enclosing boundary is shaped in a representation tailored to the data rather than fixed in advance. The review highlights how these methods translate naturally to time series once windows or learned embeddings are used as inputs. Their strength is a principled geometric formulation with well-understood optimization, but their weakness is sensitivity to the boundary’s tightness: a boundary drawn too loosely admits anomalies, while one drawn too tightly flags ordinary variation, which brings the review directly to the problem of threshold calibration.

Density-based approaches model the probability distribution of normal data and treat low-probability regions as anomalous. Kernel density estimation and local outlier factor are the classical tools, and hidden Markov models add a temporal dimension by capturing the sequence of states through which a normal process moves. The review notes that these methods excel when normal behavior has rich, multimodal structure — several distinct operating regimes, seasonal cycles, or periodic patterns — because a mixture-of-Gaussians or state-based model can assign high likelihood to each regime while penalizing transitions or values that never occur in training. The difficulty is that density estimation in high dimensions is notoriously hard, and time series derived from industrial sensors or financial markets often live in exactly such spaces. The review’s discussion of periodicity-enhanced frameworks, such as the PE-DOCC approach, illustrates how researchers inject structural knowledge about seasonality directly into the model to make the density estimation tractable and the resulting anomaly scores more meaningful.

Reconstruction-based methods, which dominate much of the modern deep learning literature, train a model — often an autoencoder, variational autoencoder, or sequence-to-sequence network — to compress and then rebuild normal time series. The guiding intuition is that a network trained exclusively on normal patterns learns to reconstruct them faithfully, but stumbles when asked to reproduce anomalous segments it has never seen, producing large reconstruction errors that serve as anomaly scores. The review catalogs numerous variants, including adversarial architectures like MAD-GAN, which pairs a generator with a discriminator to sharpen the distinction between real and generated normal data, and calibrated approaches such as COUTA that explicitly tune the decision threshold. Reconstruction methods are attractive because they handle multivariate streams with complex temporal dependencies, but the review is candid about a known pitfall: networks can generalize so well that they reconstruct even anomalous inputs accurately, muting the very signal the system is meant to detect.

The two representation-learning families reflect the field’s recent turn toward learning what to measure before learning what is normal. Feature-representation methods transform raw sequences into embeddings — using tools ranging from discrete Fourier and cosine transforms to learned signal transformation networks like OCSTN — and then apply classical one-class techniques in that transformed space. Contrastive-representation methods go further, training networks to pull similar segments of normal data together and push dissimilar ones apart, so that the resulting embedding space naturally concentrates normal behavior. Approaches such as COCA and CTAD exemplify this trend, and the review’s inclusion of boundary-driven active learning, BALAD, shows how the boundary and representation ideas are beginning to merge. The promise of these methods is robustness: a good representation can make anomalies obvious even when raw signals are noisy or nonstationary. The risk is that contrastive objectives, designed for general-purpose similarity, may not align with the specific deviations that matter in a given application.

The review’s treatment of applications reveals how widely these techniques have spread. In precision manufacturing, computer numerical control machining datasets and bearing fault benchmarks such as the Case Western Reserve University dataset test whether models can catch mechanical degradation before catastrophic failure. In healthcare, electrocardiogram, electroencephalogram, and ballistocardiography signals provide physiologically meaningful streams where labeled disease examples are scarce and patient populations vary enormously. In infrastructure and cybersecurity, the Secure Water Treatment and Water Distribution testbeds, the Soil Moisture Active Passive satellite data, the Mars Science Laboratory rover telemetry, and server machine datasets challenge models with multivariate, high-frequency streams where attacks and faults are rare by design. Financial applications, including work on the Korea Composite Stock Price Index, the S&P 500, and the Nasdaq, push the paradigm toward regimes where “normal” itself evolves continuously. The breadth of these benchmarks, curated under archives such as UCR and UEA, gives the field common ground for comparison — though the review argues that benchmarking remains far from standardized.

Indeed, the open problems the review identifies are as instructive as its taxonomy. Threshold calibration, the conversion of continuous anomaly scores into binary alarms, remains a persistent weakness, with approaches like native anomaly-based calibration and uncertainty modeling-based calibration proposed as remedies. Concept drift — the slow transformation of what counts as normal as machines wear in, patients change, or markets shift — undermines models trained on static snapshots of healthy behavior. Explainability is another gap: operators in safety-critical settings need to know not just that something is anomalous but which sensor, which frequency band, or which temporal pattern triggered the alarm, motivating explainable frameworks such as XOCTSC. Computational efficiency matters equally, since deployment on embedded controllers, satellites, or edge devices imposes memory and latency budgets that many deep architectures exceed. Evaluation practice itself is contested, with the review examining metrics from AUROC and AUPR to point-adjusted precision and average run length, each of which can paint a different picture of the same detector.

For practitioners, the review’s comparative framing offers a practical decision aid. Organizations with limited labeled data and modest compute may find distance or density methods sufficient and interpretable; those facing high-dimensional multivariate streams will likely gravitate toward reconstruction or contrastive approaches despite their heavier training requirements. The authors, based at the United Arab Emirates University and funded through university grants, position the work explicitly as a bridge between conceptual understanding and deployment insight — a recognition that the gap between benchmark performance and operational reliability is where most anomaly detection projects stumble. As sensors proliferate across industry, medicine, and finance, and as the volume of unlabeled temporal data outpaces any realistic labeling effort, the one-class paradigm the review maps is likely to move from the margins of machine learning toward its center, making this systematic synthesis a timely reference for anyone building systems that must know when something has gone wrong without ever having been shown what wrong looks like.

Subject of Research: Systematic review of time series one-class classification methods for anomaly detection in temporal data.

Article Title: Time series one class classification: a systematic review of methods, applications, and challenges

Article References: Zaitouny, A., Krishnan, A., Sherif, M., & Zaki, N. (2026). Time series one class classification: a systematic review of methods, applications, and challenges. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11692-6

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11692-6

Keywords: time series analysis, one-class classification, anomaly detection, machine learning, deep learning, autoencoders, support vector machines, contrastive learning, concept drift, industrial monitoring, cybersecurity, healthcare