International Journal of Advanced Engineering and Technology

Authors
Manar Ahmed Mohammed Alblooshi, Fatmah Rashed Mohamed Obaid Alhefeiti, Muhammed Sirajul Huda Kalathingal, Shaher Bano Mirza, Fouad Lamghari Ridouane

Abstract
Predicting the weather accurately is essential for daily life. The systematic recording of meteorological conditions is beneficial to many fields, but particularly to agriculture and other industries that rely on it. The overarching goals of this research are to develop an accurate and flexible statistical model for making city-level weather predictions and assess how intermediate meteorological variables affect the LSTM model's performance. Success with this method on many different challenging prediction issues can be attributed to the sophistication of modern neural networks. The LSTM model's results for weather forecasting are plotted using the standard Python library matplotlib. pyplot. Artificial neural networks with deep layers can model complex data structures. In order to control the data stored in a cell state, Long Short-Term Memory (LSTM) units implement input, output, and forget gates. The error rates of LSTM models are lower than those of other models, allowing them to be used more frequently for predicting. In our analysis, stacked LSTM outperformed both single-cell and bidirectional models when making predictions. The mean square error with single-cell models has an average value of 0.52 based on the weather data we have. The bidirectional approach achieved an astounding 0.99 with the lowest error rate of 0.3 for temperature accuracy, and similarly impressive results were achieved for wind speed accuracy. However, it recorded a low accuracy rating of 0.84 with 31.23 MSE while trying to gauge humidity.