Performance Analysis of Deep Neural Network Models for Weather Forecasting in Bangladesh

Abstract

Climatology and Weather forecasting play an important role to determine future climate expectations and help the farmer to make a plan for crop irrigation, fertilization, and suitable days for working in the field. Forecasting weather is a challenging task due to the uncontrolled nature of the surrounding atmosphere. Nowadays, deep learning models are widely used for weather forecasting that explores the hidden hierarchical patterns in big weather datasets to extract high-level features. In this paper, we investigate the performance of Multilayer Perceptron (MLP), ARIMA, and Bi-directional Long Short-Term Memory (BiLSTM) models for forecasting weather in agricultural applications of Bangladesh. The performance of the models is evaluated using Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). The results depict that BiLSTM provides better performance compared to other state-of-the-art models to predict accurate weather in Bangladesh.