Comparative Analysis of LSTM, BILSTM and ARIMA for Time Series Forecasting on 116 years of Temperature and Rainfall Data from Pakistan

Abstract

Numerous aspects of human life, including agriculture, transportation, and health, are significantly influenced by weather, both economically and socially. Rain has an impact on landslides, floods, and other natural disasters. We are motivated to create a model for comprehending and forecasting rain in order to provide advanced warning in a spectrum of areas such as transport, agriculture, and so on because of the numerous consequences that rain and temperature have on human survival. In this study, a dataset for temperature and rainfall for Pakistan for 116 years is used. Comparative analysis of ARIMA, LSTM and BILSTM is performed. For this study, 90% of data is used for training and the 10% for testing. Normalization is also performed to clean data. According to the results, LSTM and BILSTM are better than ARIMA but for specific cases of rainfall, BILSTM performed better than LSTM and for Temperature LSTM outperformed BILSTM.