A hybrid SVR, XGBoost and LSTM model for temperature forecasting in Telangana using ERA5 reanalysis data

It is important to have proper weather prediction in climate sensitive areas such as Telangana in India. The work is a comparison of the SVR, XGBoost, and LSTM models for predicting daily maximum temperature (Tmax) using ERA5 reanalysis of data (2010 - 2025) comprising eight normalized weather parameters. The individual accuracy of SVR was highest (RMSE = 0.10 °C, MAE = 0.07 °C, R 2 = 1.00), with XGBoost second (RMSE = 0.14 °C, MAE = 0.07 °C, R2= 1.00). LSTM (RMSE = 1.85 °C, MAE = 1.46°C, R2= 0.50) was shown to capture time trends nicely and less accurate in general. A Hybrid model was established to strengthen performance which was the synthesis of the outputs of all the three models through the process of weighted averaging. It provided the most accurate results (RMSE=0.07 °C, MAE=0.05 °C, R2=1.00) proving that the combination of machine learning and deep learning approaches contributes to more accurate prediction of regional temperatures.