Machine Learning Model Performance in Dry and Wet Seasons

Historical data of daily volume of water consumed in dry and wet seasons was gathered through questionnaires, and integrated with socioeconomic data, weather data, property data and geospatial data. The datasets were passed through Principal Component Analysis algorithm to select few features that explain the variability in all the original variables.

The selected features were inputted into four predictive models – Multilinear Regression (MLR), Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANN). Three error metrics, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and R squared (R2) score were used to measure the model performances. All four models performed very well in predicting volume of water consumed by poor urban households as they produced RMSE of 57 litres, 46 litres, 63 litres and 52 litres respectively during training, and R2 score of 88%, 92%, 88% and 90% respectively.