Air Quality Prediction: A Case of Indian Cities

Classical machine learning algorithms have shown success in predicting the Air Quality Index (AQI) for Indian cities, but many previous approaches overlook the temporal dynamics of air quality data, leading to overly optimistic results that fail in real-world applications. This paper presents an alternative approach that incorporates lagged pollutant values and employs nested time series crossvalidation to provide a more accurate and robust AQI prediction across different cities using various regression models. Boosting algorithms, in particular, demonstrated superior performance, with Gradient Boosting achieving an R² score of 94.24% for Delhi. The study concludes that boosted decision trees effectively capture temporal dependencies, showing strong potential for real-life AQI applications as more data becomes available.