Adaptive Temperature Prediction Using Incremental Learning: Enhancing Intelligent Computing for Climate Sustainability

Global warming and changing climate dynamics are causing extreme weather events such as heat waves. These extreme weather events are occurring more frequently with more intensity and demand the urgent need for accurate and adaptive temperature prediction methods. Conventional techniques are usually based on static models that are not capable of adapting to new data and changing patterns in real-life. These models require retraining from scratch on new data and this restricts their performance in dynamic and changing climates. In this study, we propose an incremental learning-based approach for temperature prediction, with the help of spatio-temporal data from India Meteorological Department (IMD) for the period 2014-2023. The dataset is divided into seven climatic regions and four seasons. Our solution employs Long Short-Term Memory (LSTM), Random Forest, and Naïve Bayes classifier. The data, which consists of daily records across a 31x31 spatial grid, is preprocessed using MinMaxScaler. Incremental learning techniques are used to handle the large data size by processing it in manageable chunks. The models are evaluated using key metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R-squared (R2), and Mean Absolute Percentage Error (MAPE). The incremental LSTM model outperformed the others, achieving the highest predictive accuracy with an R2 score of 0.94. These results highlight the importance of incremental learning approaches for large-scale climatic data.