تصنيف مستويات أداء الشركات المدرجة في سوق دمشق للأوراق المالية والتنبؤ بها باستخدام التحليل العنقودي الهرمي وشبكة NN RBF

Abstract

This study aims to evaluate and classify the performance of companies listed on the Damascus Securities Exchange by integrating two complementary approaches: hierarchical cluster analysis as an unsupervised method to uncover latent performance structures, and a radial basis function neural network (NN-RBF) as a supervised method for performance prediction and generalization assessment. The analysis relies on market and financial indicators capturing trading dynamics, investor attention, and profitability, including returns, number of transactions, stock turnover ratio, number of trading days, number of investors, and earnings per share, with standardized inputs to ensure measurement comparability. The clustering results identify four performance categories—low, moderate, good, and excellent—while silhouette coefficient (which is considered efficient tool to identify optimal number of groups) indicates stronger cohesion for the middle categories and greater overlap at the extremes, reflecting structural heterogeneity in market behavior. For the predictive stage, the model yields stable performance, as indicated by closely aligned error rates across training and testing. Multi-class ROC analysis further confirms strong discriminatory power, particularly for the excellent and low performance classes. Variable-importance findings reveal that the number of investors, followed by the number of transactions, exerts the greatest influence on performance classification, underscoring the economic role of market depth and liquidity in improving firms’ positioning into higher performance categories