Deep Learning Web Based Prediction System for Zakat Collection

Abstract

Zakat is a religious obligation that aims for charitable contribution by the Muslims. It is an annual payment made under Islamic law on certain kinds of property by a group of Muslim and is based on certain criteria. The collected zakat is distributed to certain group people based on pre-defined criteria with the aim to improve their life well-being. The payment and distribution of zakat have played a major role in the history of Islam. To date, various studies have been reported on utilizing ICT, in particular deep learning models, to facilitate either the zakat collection or distribution. However, unlike artificial neural network, deep learning consists of more hidden layers. One of the issues in the deployment of deep learning models is the determination of number of hidden layers and hidden neurons. This study demonstrates the use of Long Short Term Memory (LSTM), to forecast zakat collection for one of the states in Malaysia. The model is realized through four phases: data collection and preparation, algorithm design, model implementation and model evaluation. The focus was put in investigating the effects of LSTM parameters (window size, number of hidden layers and number of hidden nodes) on the zakat forecasting performance.  Experimental results showed that LSTM produces the least error rate (i.e 5.29) when it uses two hidden layers, three hidden nodes and seven time step (i.e window size). With an error rate below than 8%, the produced forecasting model can be considered as a competitive prediction model. Putting the prediction model into practice, zakat institution will be able to better design appropriate strategies to collect and distribute zakat for the well-being of the ummah.