Predicting Economic Activities Trends for Saudi and Non- Saudi Establishments in Saudi Arabia by Using Time Series Forecasting and Deep Learning Models

Abstract

Data related to economic and stock market information are widely used for representing non- stationary time series. The economy goes through different phases, from recession to growth. This project focused on models predicting total added establishments in each economic activity. It aimed to predict the number of establishments after the impact of covid- 19 and estimate the accuracy of the prediction models. Prediction models used for inexpensive, quick evaluation of the added number of existing establishments leading to business risk mitigation. The research was conducted on data extracted from ministry of human resource and social development database. Therefore, the carried- out analysis highlights four activities under examination to capture the significance changes. These activities are construction, health, accommodation and information technology. In this study, five time series models are selected and applied to the business data describing expansion and recession. After model performance evaluation, deep learning models with respect to sliding window approach to predict short term values are recommended and perform better than traditional models. LSTM outperforms the other models in health with 18.22 RMSE and 90.65 RMSE for the information technology. DNN with two hidden layers got the best RMSE for accommodation and construction activities which is equal to 183.98 and 1387.78 respectively. Such work indicates that predicting overall added establishments may assist investors and companies in making economic choices, such as when to invest, increase, or reduce production.