Ph.D. dropout is a persistent and challenging issue in higher education, with significant implications for individual students, academic institutions, and the broader society. This research paper aims to explore the factors contributing to Ph.D. students' decision to dropout and their interrelationships. For this purpose, we employed the hybrid topic modeling Bidirectional Encoder Representations from Transformers -- Latent Dirichlet Allocation (BERT-LDA) algorithm and Numerical Association Rule Mining (NARM) using a genetic algorithm in QuantMiner. We identified and analyzed individual, institutional, and social factors that affect Ph.D. students in leaving their current degrees. The results suggest that financial constraints, inadequate academic preparation, poor mentoring, social isolation, lack of social support, family responsibilities, and work-life balance are significant elements responsible for dropout. These findings also reveal that these factors are interrelated, and their effects can be mitigated by the academic institution's policies and culture. The outcomes of the study have implications for academic institutions, policymakers, and researchers, who can use them to develop evidence-based strategies and interventions that enhance Ph.D. students' retention and success.