Hybrid Machine Learning–Deep Learning Framework for Time-Series Rainfall Prediction in Tamil Nadu

Abstract

Accurate rainfall prediction is essential for agricultural planning, irrigation scheduling, water resource management, and disaster mitigation in Tamil Nadu, a region strongly influenced by Southwest and Northeast monsoon systems. This study proposes a converged Machine Learning (ML) and Deep Learning (DL) framework for time-series rainfall prediction using historical district-wise monthly rainfall data collected from the India Meteorological Department (IMD) for the period 2000–2024. The dataset includes rainfall records along with meteorological parameters such as temperature, humidity, wind speed, and seasonal indicators from selected Tamil Nadu districts including Chennai, Coimbatore, Madurai, Tirunelveli, and Cuddalore.
The proposed framework integrates Random Forest (RF) and Long Short-Term Memory (LSTM) networks to improve rainfall forecasting accuracy. In the preprocessing stage, missing values were handled, features were normalized, and temporal lag features (1–12 months) were generated to capture seasonality and rainfall trends. Random Forest was employed for feature selection and prediction refinement, while the LSTM model captured long-term temporal dependencies in rainfall patterns. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE).
Experimental results demonstrate that the hybrid ML–DL model outperformed standalone ML and DL models, achieving an RMSE of 16.82 mm, MAE of 12.94 mm, and MAPE of 10.12%. The proposed approach effectively captured seasonal rainfall variations and extreme monsoon events across different districts. The findings indicate that the hybrid framework can support district-level weather forecasting and climate-resilient decision support systems for agriculture and water resource management in Tamil Nadu..