استخدام تقنيات التعلم الآلي لتحليل البيانات الاقتصادية في سورية لدعم القرارات الاقتصادية

Abstract

This research explores the potential of applying machine learning and data analysis techniques to support economic decision-making in Syria, particularly in light of the economic challenges and disruptions facing the country. The research highlights the limitations of traditional economic models based on the Maximum Likelihood Estimation (MLE) method and emphasizes the ability of machine learning to handle big data and capture complex patterns, thereby providing deeper and more accurate insights into economic indicators and trends.

The research addresses three main machine learning techniques and their applications to Syrian economic data: First, it demonstrates how neural networks, including Neural Network Autoregression (NNAR) models and Gated Recurrent Unit (GRU), can be used to forecast economic variables such as inflation rates and the Damascus Securities Exchange Index. Second, it illustrates the use of random forests to analyze textual data extracted from Google Trends, aiming to predict investment decisions in the Damascus Securities Exchange based on search patterns for keywords related to the Syrian economy. Third, the research tackles the challenge of prediction in a high-dimensional data environment and proposes the use of Lasso regression with L1 penalty to shrink coefficients and select the most influential variables on the Consumer Price Index (CPI), enhancing prediction accuracy and reducing variance. The research concludes that machine learning techniques are effective tools for analyzing complex Syrian economic data and provide valuable insights for supporting economic decision-making. It recommends further integration of these techniques into economic planning and analysis processes in Syria to develop more accurate predictive models, contributing to more informed and effective policy decisions and helping to mitigate economic uncertainty