Long Short Term Memory and Non-linear Autoregressive Models for Prayer Time Prediction

since the advancement of Artificial Neural Network (ANNs), many studies have been proposed in the past few decades for forecasting purposes. Time-series applications have been applied in areas such as stocks, electricity, and weather prediction. The role of time in conducting prayer cannot be over-emphasized. Muslims must perform each prayer at its stipulated time for it to be valid. But, overall these years, nothing has been done yet about prayer time prediction. In this paper, the Long Short-Term Memory (LSTM) of Recurrent Neural Network (RNN) models and Non-linear Autoregressive (NAR) models are utilized to exploit the use of time series application in Kano state prayer time prediction. The models mainly focus on collecting Kano state prayer times from prior years to predict future prayer times for three prayers (noon prayer, sunset prayer, and evening prayer). The LSTM training algorithm for the LSTM model, the Bayesian Regularization Neural Network (BRNN) algorithm, and the Levenberg-Marquardt (LM) algorithm for the NAR model were the state-of-the-art algorithms used in training the models. After that, a comprehensive comparative analysis is conducted between the algorithms to determine the best performing algorithm. Finally, the models were tested with a dataset from the Kano state shari'ah commission for prayer time from 2013–2020, for three prayers (noon prayer, sunset prayer, and evening prayer). The accuracy of the result shows that LSTM and NAR models could be applied to prayer time prediction and obtain decent accuracy.