Rainfall Prediction Using Machine Learning: LSTM

Abstract

The research problem being proposed is to determine how the Long Short-Term Memory (LSTM) networks forecast daily rain in the Sylhet district, Bangladesh, between June 2014 and June 2023. LMST is a form of recurrent neural network (RNN), which was utilized because it is capable of modelling the temporal association of the sequential information. The model has been trained based on data between 2014 and 2021, and the predictions have been tested up to 2022 and 2023. The results indicated that the LSTM achieved a total accuracy score of 82 with a high-performance rate at the monsoon approaching normal season (May to September). The model, however, presented some limitations in forecasting occasional rainfall events in the dry and transition months. The paper indicates the possible use of LSTM networks in the prediction of rainfall, particularly where forecasting roles of weather vary significantly. Further development could be achieved through the inclusion of more meteorological variables, improvement of data resolution, and also consideration of a hybrid form of modelling to determine predictions of extreme precipitation or rainfall predictions more perfectly.