Rainfall prediction system for Bangladesh using long short-term memory

Mustain Billah, Md. Nasim Adnan, Mostafijur Rahman Akhond, Romana Rahman Ema, Md. Alam Hossain, Syed Md. Galib

Open Computer Science · 01 January 2022

In one sentence. Two-layer LSTM predicts Bangladesh rainfall from 10 selected features with 97.14% accuracy, outperforming conventional ML baselines; the evaluation is limited.

Abstract

Abstract Rainfall prediction is a challenging task and has extreme significance in weather forecasting. Accurate rainfall prediction can play a great role in agricultural, aviation, natural phenomenon, flood, construction, transport, etc. Weather or climate is assumed to be one of the most complex systems. Again, chaos, also called as “butterfly effect,” limits our ability to make weather predictable. So, it is not easy to predict rainfall by conventional machine learning approaches. However, several kinds of research have been proposed to predict rainfall by using different computational methods. To accomplish chaotic rainfall prediction system for Bangladesh, in this study, historical data set-driven long short term memory (LSTM) networks method has been used, which overcomes the complexities and chaos-related problems faced by other approaches. The proposed method has three principal phases: (i) The most useful 10 features are chosen from 20 data attributes. (ii) After that, a two-layer LSTM model is designed. (iii) Both conventional machine learning approaches and recent works are compared with the LSTM model. This approach has gained 97.14% accuracy in predicting rainfall (in millimeters), which outperforms the state-of-the-art solutions. Also, this work is a pioneer work to the rainfall prediction system for Bangladesh.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.