Evaluating Rainfall Prediction Models on Time Series Data from Bangladesh and Pakistan: A Comparative Approach

Abstract

Predicting rainfall is obviously a challenging task due to the multitude of factors and elements that impact climate conditions. Accurate rainfall forecasts are extremely important, especially for the agriculture industry, which depends heavily on timely and adequate rainfall for crop growth and result. The need for precise rainfall forecasts is further highlighted by the economic contribution that agriculture makes. Around the world, a variety of techniques have been employed to forecast rainfall patterns. This paper presents a comparative analysis of various rainfall prediction models, including statistical, machine learning on time series data and statistical approaches. We evaluate the performance of ARIMA, Random Forest, Linear Regression, Gradient Boosting and SVM models on diverse datasets from different geographical regions and climatic conditions. This study evaluates rainfall prediction using both machine learning and statistical model on distinct datasets from Bangladesh and Pakistan. The ARIMA (3, 2, and 1) model is applied to both datasets, demonstrating consistent performance on the Pakistan dataset with minimal differences between in-sample and out-of-sample results, indicating reliable forecasting ability. In contrast, the Bangladesh dataset shows a noticeable drop in performance from in-sample to out-of-sample, suggesting potential over fitting. Additionally, machine learning models such as Gradient Boosting (GB), Random Forest (RF), Support Vector Regressor (SVR), and Linear Regression (LR) are utilized. For the Bangladesh dataset, Gradient Boosting outperformed others with the lowest error values (MSE: 6435.975, RMSE: 80.225, MAE: 51.064) and the highest R² score (0.842). On the Pakistan dataset, Linear Regression produced the best results with the lowest MSE (270.138), RMSE (16.436), and MAE (11.572).Our findings highlight the strengths and limitations of each model, offering insights into their applicability for accurate rainfall forecasting.