Evaluation of Machine-Learning Algorithm’s Skill for Convective Weather Forecasting in Uttarakhand

This research focuses on developing a rainfall prediction model using a polynomial regression approach with nine weather variables as weather indices. The dataset comprises 2806 values for each weather variable obtained from the Indian summer monsoon of June to September months over 23 years (2000–2022). Performance evaluation of the polynomial regression model is conducted using metrics such as RMSE, MAE, MSE, and R2, along with cross-validation techniques to determine the optimal polynomial order and prevent overfitting. The results indicate that the second-degree polynomial regression model demonstrates the best accuracy and precision in forecasting rainfall. Scatter plots and residual distribution analysis further confirm the model's effectiveness in predicting rainfall for different categories. Additionally, forecast skill analysis shows satisfactory correlations between observed and predicted rainfall, especially for light rain and medium rain categories. This study contributes valuable insights into enhancing weather forecasting models for rainfall prediction.