Machine Learning Based Analysis of the Impact of University Specialization on Unemployment Rate رسالة ماجستير

Abstract

Palestine, like most developing and Arab countries, suffers from the problem of unemployment. This problem in the Palestinian economy is linked to many economic and social variables, especially growth. As in 2022, the unemployment rate among those participating in the labor force in Palestine stood at 24.4%, compared with 23.0% in 2015. Moreover, a significant percentage of Palestinian youths and graduated face limited job prospects, leading to frustration and disillusionment. High levels of youth unemployment not only jeopardize individual opportunities but also pose a broader societal challenge, potentially contributing to social unrest. According to figures from the Palestinian Central Bureau of Statistics (PCBS), the unemployment rate for individuals aged 20 to 29 with an intermediate diploma, bachelor's degree, or higher has reached more than 48%. This translates to half of the graduates being unemployed. Given the importance of the issue of unemployment among graduates in Palestine, in this research we learned about the reality of unemployment among graduates, and worked to apply machine learning techniques to classify labor force data to determine the reality of young graduates in Palestine in terms of work and unemployment, and determine the impact of some variables on employment status such as (gender, age, specializations university, region, governorate,...), as unemployment deserves special attention among university graduates and their field of study. Two sets of classification models were built: one with five algorithms (Random forest (RF), Decision tree (DT), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors Algorithm (KNN), Adaptive Boosting (AdaBoost)) for all graduate data, and another with three algorithms (RF, DT, and KNN) for data grouped by university major. RF achieved an impressive 95.8% accuracy on the test data, DT was close behind with 93.4%. Other models like XGBoost, KNN, and AdaBoost performed well too, with accuracies 91.6, 88% and 91.8% respectively. Also the RF excelled in specialization classification, achieving impressive accuracy across all four categories: business and administration and law (94.9%), engineering, manufacturing and construction (94.7%), education (94.1%), and health & welfare being (94%).