Predicting Air Quality Index in Accra using Machine Learning

Air quality is a significant public health issue, and accurate predictions of the Air Quality Index (AQI) are crucial for timely interventions. This study explores the use of supervised machine learning algorithms to forecast AQI across different neighborhoods in Accra, Ghana. Six models including Random Forest , CatBoost, Support Vector Regression (SVR), Linear Regression, Ridge and Lasso Regressions, were evaluated. Data from the Breathe Accra platform, encompassing pollutants and weather conditions in five neighborhoods, were preprocessed through data cleaning, feature selection, and normalization. Model performance was assessed using the coefficient of determination (R ² ), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). While the initial performance of the models was less satisfactory, the incorporation of wavelet transform preprocessing significantly enhanced the results. This improvement was particularly notable for the area of Korle Bu, where the CatBoost model's R ² increased from 0.32 to 0.58, RMSE decreased from 29.03 to 22.79, and MAE dropped from 22.11 to 17.01. On average, all models except SVR performed well across all areas, as evidenced by the evaluation metrics. These findings have direct implications for enhancing air quality management and policymaking in the city of Accra, where accurate AQI predictions are vital for effective public health interventions and environmental planning.