The significance of the result lies less in the headline accuracy figure than in what it implies for the millions of people living in rapidly urbanizing regions where air quality monitoring infrastructure is thin. The Air Quality Index, or AQI, condenses the concentrations of several regulated pollutants into a single number that public health authorities use to issue warnings, close schools, and restrict traffic. Producing that number reliably normally requires continuous measurements of particulate matter, sulfur dioxide, carbon monoxide, and nitrogen dioxide from calibrated instruments. When a district has only sparse or intermittent monitoring, the index becomes unreliable exactly when residents need it most, and the health consequences of missed pollution episodes can be severe.

Akiner and Ghasri attacked the problem with a two-stage design that reflects a key insight about atmospheric data: pollution does not behave the same way under all weather conditions. In the first stage, they applied K-Means clustering, an unsupervised algorithm that groups observations by similarity without being told what the groups should be, to a combination of meteorological variables and pollutant concentrations. The algorithm partitioned the ten-year record into distinct atmospheric regimes, essentially discovering recurring weather-and-pollution patterns on its own. Each cluster corresponds to a characteristic set of conditions, such as stagnant, cold episodes in which pollutants accumulate near the surface, or windy, well-mixed periods in which concentrations disperse rapidly.

This clustering step matters because it gives the downstream model a structured picture of the atmosphere instead of forcing it to learn every relationship from raw, undifferentiated data. In the second stage, the researchers employed XGBoost, a gradient-boosted decision tree algorithm that has become a workhorse of tabular machine learning. XGBoost builds an ensemble of decision trees sequentially, with each new tree trained to correct the residual errors of the ensemble so far, and it includes regularization terms that discourage overfitting. The model was tasked with estimating AQI values from meteorological parameters together with the key pollutants PM10, PM2.5, SO2, CO, and NO2, conditioned on the regime structure identified by the clustering stage.

The crucial design choice is the framework’s reliance on widely available meteorological data to compensate for scarce pollutant measurements. Weather variables such as temperature, wind speed, humidity, and pressure are collected routinely by national meteorological services almost everywhere on Earth, and Türkiye’s General Directorate of Meteorology, together with the Ministry of Environment, Urbanization, and Climate Change, provided the database underpinning the study. Because these inputs are cheap and abundant, the framework reduces dependence on dense sensor infrastructure. In effect, the model learns how the local climate converts emissions into pollution episodes, and can then reconstruct or forecast air quality even where direct pollutant observations are limited.

The reported performance is remarkable for a real-world urban dataset. An overall accuracy of approximately 98 percent means the model reproduces the observed AQI with enough fidelity to support timely health advisories, giving authorities lead time to warn vulnerable populations, adjust traffic, or deploy inspection resources. Beyond raw accuracy, the authors used feature-importance analysis to identify which meteorological drivers dominate AQI variability in Kağıthane. This kind of interpretability is increasingly demanded of machine learning in environmental applications, because a black-box prediction is of limited use to a planner who needs to know whether ventilation, temperature inversions, or precipitation scavenging is controlling pollution on a given day.

Kağıthane itself is a fitting laboratory for the method. The district sits in a valley that channels both traffic and terrain-driven airflow, and previous research has documented elevated particulate levels in the Kağıthane Valley and the wider Istanbul metropolitan area. Istanbul combines intense vehicular emissions, a population exceeding fifteen million, complex topography, and a coastal climate whose winds and humidity strongly modulate pollutant dispersion. A model that performs well in such a demanding setting is more likely to transfer to other challenging environments than one tuned on a flat, well-monitored city.

The authors emphasize that the framework is modular and readily transferable across different geographic and temporal contexts. Both K-Means and XGBoost are standard, well-documented algorithms, and the input data are commonly available, which means a municipality does not need supercomputing resources or proprietary technology to adopt the approach. The pipeline can be retrained on local data, the clustering stage will rediscover the atmospheric regimes specific to a new location, and the boosted trees will re-weight the meteorological drivers accordingly. That scalability is the study’s central claim: a practical answer for regions facing environmental monitoring constraints, from fast-growing cities in the Global South to districts of wealthy cities where sensor coverage is uneven.

The work also arrives amid a broader shift in how air pollution is measured and modeled. Researchers worldwide are experimenting with low-cost sensor networks, satellite retrievals of aerosol optical depth, and deep learning architectures for spatiotemporal forecasting, and systematic reviews have catalogued both the promise and the pitfalls of machine learning in air pollution epidemiology. Against that backdrop, the Istanbul study makes a deliberately pragmatic argument: rather than waiting for perfect instrumentation, cities can extract far more value from the data they already possess by combining unsupervised structure discovery with powerful supervised regression. The approach complements, rather than replaces, physical dispersion models, which remain essential for source attribution and regulatory analysis.

For public health, the implications are immediate. Air pollution is linked to cardiovascular and respiratory disease, adverse birth outcomes, and growing evidence of effects on mental health and children’s cognitive development, and the burden falls disproportionately on neighborhoods with the least monitoring. A framework that turns routine weather data into near-accurate AQI forecasts can level that playing field, enabling evidence-based environmental policy in places that have historically flown blind. As urbanization accelerates and climate change alters the meteorological conditions that govern pollutant dispersion, tools like this hybrid model may become as standard a piece of civic infrastructure as the weather forecast itself, quietly translating the atmosphere’s behavior into actionable warnings for the people who live beneath it.

Subject of Research: Hybrid machine learning for air quality index forecasting in data-scarce urban environments

Article Title: Overcoming data shortages through hybrid machine learning in air quality forecasting: a case study of Kağıthane, Istanbul

Article References: Akiner, M. E., & Ghasri, M. (2026). Overcoming data shortages through hybrid machine learning in air quality forecasting: a case study of Kağıthane, Istanbul. Theoretical and Applied Climatology, 157(10), Article 609. https://doi.org/10.1007/s00704-026-06545-9

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06545-9

Keywords: air quality, machine learning, XGBoost, K-Means clustering, Air Quality Index, Istanbul, Kağıthane, meteorology, urban pollution, data scarcity, public health, environmental forecasting

The research team, led by Priyal Singhal and Upendra Singh of Manipal University Jaipur together with colleagues at VIT Bhopal University, trained and validated their framework on 435,742 daily air quality records collected by India’s Central Pollution Control Board between 1990 and 2015. The dataset spans monitoring stations across major states including Maharashtra, Uttar Pradesh, Andhra Pradesh, Punjab, and Rajasthan, recording concentrations of sulphur dioxide, nitrogen dioxide, respirable suspended particulate matter, and suspended particulate matter. Because the historical records predate routine measurement of PM2.5, carbon monoxide, ozone, ammonia, and lead at most stations, the researchers computed a CPCB-derived AQI proxy based on the three pollutants reliably available throughout the period, with particulate matter serving as the historical Indian monitoring proxy for PM10.

The first stage of the pipeline addresses the missing data problem, which is far from random in this dataset. Gaps correlate with station age and with the monsoon months of June through September, when outages are most frequent. Rather than deploying computationally expensive generative adversarial networks such as TimeGAN, the team developed a lightweight noise-augmented linear interpolation scheme. The method fills gaps by interpolating between observed neighbours, then injects calibrated Gaussian noise with a standard deviation set to two percent of each column’s observed dispersion, preventing the artificially smooth segments that plague simple interpolation. A final clipping step enforces physical non-negativity. The approach comes with a formal variance-preservation guarantee: the ratio of imputed to observed variance stayed within 0.04 percent of unity across pollutants, and sensitivity analysis confirmed that performance is stable for noise scales between 0.01 and 0.05. Compared against mean-fill, forward-fill, KNN, MICE, Kalman smoothing, and MissForest under an identical downstream pipeline, the scheme delivered the lowest error at fifteen to twenty times lower computational cost than the heaviest alternatives.

At the heart of the framework sits a hybrid ensemble combining two fundamentally different modelling philosophies. A long short-term memory network with a fourteen-day input window learns sequential dependencies, capturing fortnight-scale dynamics such as multi-day pollution accumulation and monsoon wash-out episodes that discrete lag features cannot fully encode. In parallel, an XGBoost gradient-boosted tree model exploits nonlinear feature interactions across twenty-six engineered predictors, including lagged values, rolling means, calendar variables, and inter-pollutant ratios. The two sub-models are fused through a dynamic R-squared-weighted averaging scheme derived from a variance-minimising convex combination argument. The measured residual correlation between the two learners was just 0.31, indicating genuine complementarity, and covariance-optimal weights estimated independently fell within 0.04 of the R-squared-derived weights, confirming the fusion rule operates near the theoretical optimum.

The most distinctive contribution, however, is the closed-loop use of explainability. Rather than computing SHAP values merely for human inspection, as previous LSTM-SHAP frameworks have done, the team used SHAP attributions from the XGBoost sub-model as an active feature-selection signal. SHAP, grounded in cooperative game theory, assigns each feature an exact marginal contribution to every prediction, and for tree models the TreeExplainer algorithm computes these values in polynomial time. The researchers independently verified feature importance for the LSTM using gradient sensitivity analysis, finding strong agreement between the two methods: Spearman rank correlation of 0.87, with the top three features identical across both approaches. The top ten features by SHAP importance were then used to retrain both sub-models, cutting input dimensionality by 62 percent.

The results are striking. The final XAI-enhanced ensemble achieved a root mean squared error of 0.3817, a mean absolute error of 0.2773, and an R-squared of 0.9994 on the held-out test set, representing a 54.7 percent RMSE reduction and a 58.7 percent MAE reduction over the standalone LSTM baseline. Translated to original AQI units, the error corresponds to roughly 1.91 AQI points, well below the 50-point width of the narrowest CPCB AQI category and at or below the noise floor of the underlying measurement instruments, whose typical uncertainty propagates to roughly three to four AQI units. SHAP-guided selection alone yielded an additional 38 percent RMSE improvement over the standard ensemble, and the across-trial variance of test error fell by 56 percent, an effect the authors interpret through bias-variance theory: removing low-signal features suppresses estimator variance without materially increasing bias.

The team subjected these gains to unusually rigorous statistical scrutiny. All models were retrained across thirty independent random seeds, with paired t-tests and Wilcoxon signed-rank tests applied under Bonferroni correction. Every improvement remained significant by several orders of magnitude, with Cohen’s d effect sizes exceeding 2.8, far above the conventional threshold of 0.8 for a large effect. A five-fold rolling-origin time-series cross-validation, in which each fold advances the train-test boundary forward in time, reproduced the model ranking with RMSE of 0.3928 versus 0.6382 for the standard ensemble, and bootstrap confidence intervals for the two models did not overlap. The framework also outperformed a Temporal Fusion Transformer, which the authors attribute to the limited training scale under-utilising the transformer’s capacity while the ensemble’s explicit feature engineering exploits the available signal more efficiently.

The authors are candid about the caveats. The near-deterministic correlation of 0.99 between RSPM and the AQI target means that part of the headline R-squared reflects a definitional relationship, though control experiments demonstrate genuine learning: a deterministic baseline computing AQI directly from RSPM achieves substantially higher error, and removing RSPM from the feature set entirely still yields an R-squared of 0.9851. Winter pollution peaks are systematically under-predicted by four to eight AQI units, a bias traced to the absence of meteorological covariates such as temperature, wind speed, and boundary layer height in the historical records, with stubble-burning episodes in the Indo-Gangetic Plain contributing an episodic component that a seasonal correction only partially recovers. The test set also contains few samples in the Poor, Very Poor, and Severe AQI categories, precisely those most relevant for public health intervention, and the framework’s calibration to Indian RSPM-dominated profiles would require adaptation in PM2.5-dominated or ozone-dominated regions.

On the deployment front, the numbers are compelling. Per-sample inference takes just 7.9 milliseconds on a commodity CPU, roughly five orders of magnitude faster than the CPCB’s reporting cadence, and the reduced ten-feature input lowers memory and input-output demands. The explainability stage adds about twenty-one minutes at training time only. Because the pipeline is agnostic to pollutant count, porting it to post-2015 CPCB data with the full eight-pollutant suite requires only substituting the sub-index inputs and re-running SHAP selection. The study’s central lesson may prove its most enduring contribution: interpretability and accuracy are not competing objectives in environmental forecasting but complementary ones, and the feedback loop that turns explanations into better models offers a template that extends well beyond air quality, from rainfall prediction to any domain where high-stakes decisions demand both precision and justification.

Subject of Research: Explainable hybrid deep learning for air quality index prediction using Indian CPCB monitoring data

Article Title: Explainable hybrid deep learning framework for air quality index prediction in Indian cities

Article References: Singhal, P., Saurabh, P., Sharma, M., Badge, J., & Singh, U. (2026). Explainable hybrid deep learning framework for air quality index prediction in Indian cities. Discover Artificial Intelligence, 6(1), Article 1395. https://doi.org/10.1007/s44163-026-02131-0

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02131-0

Keywords: air quality index, explainable AI, SHAP, LSTM, XGBoost, ensemble learning, deep learning, missing data imputation, CPCB India, air pollution, environmental forecasting, machine learning