Analysis of Global Air-pollution Data 2025 By ML and Python

Air pollution poses a critical global health challenge requiring accurate prediction systems. This paper evaluates eleven machine learning models for air quality classification using the Global Air Pollution 2025 dataset. We compare traditional algorithms (Logistic Regression, Decision Trees, SVM) against ensemble methods (Random Forest, Gradient Boosting, XGBoost) and deep learning (LSTM) using multiple performance metrics. Statistical validation through ANOVA and Tukey’s HSD tests (F-statistic: 96.0243, p<0.0001) confirms that ensemble methods significantly outperform other approaches, with Decision Tree, Gradient Boosting, and Random Forest showing no statistically significant differences among themselves (p>0.05). SHAP analysis identifies PM2.5 concentration as the most influential predictor. The LSTM model underperformed (F1-score: 0.127) compared to traditional approaches, suggesting limitations in temporal pattern recognition with the available data. This study provides statistically validated insights for implementing effective air quality monitoring systems to support environmental policy decisions.