Comparison of Machine Learning Models for Urban Air Quality Index Prediction: A Case Study of Kadıköy District in Istanbul

This study aims to predict Air Quality Index (AQI) values by analyzing the relationship between major air pollutants (PM10, NO2, CO, SO2, and O3) and meteorological variables (temperature and wind speed) in Kadıköy, one of the most densely populated and traffic-intensive districts of Istanbul. For this purpose, three supervised machine learning algorithms, namely Decision Tree (DT), Random Forest (RF), and Multi-Layer Perceptron Neural Network (MLPNN), were employed. Hyperparameter optimization was applied to DT and MLPNN models, and all models were evaluated using k-fold cross-validation (k=5). The results indicate that RF and the optimized DT models achieved high prediction accuracy. The RF model attained a test R2 score of 93.3% and a MAPE of 2.48%, while the tuned DT model reached a slightly higher R2 of 94.0% on the test set. However, RF demonstrated more stable performance in cross-validation averages. Although MLPNN models yielded competitive results, their overall performance was lower than the tree-based models. The findings suggest that tree-based approaches provide more generalizable and robust outputs than neural models for AQI prediction. This study offers a data-driven and practical framework to support urban air quality management and environmental decision-making processes.