The Doctoral School in Science and Engineering is happy to invite you to Xinlin WANG’s defence entitled
Advanced Machine Learning Techniques for Enhancing Company Financial Health Prediction
Supervisor: Dr. Mats Håkan BRORSSONE
Small and medium-sized enterprises (SMEs) are critical to global economic stability, yet they are particularly vulnerable to financial risks and bankruptcy. This dissertation focuses on enhancing SME financial risk prediction through advanced data-driven methods. Leveraging financial and non-financial datasets, this research aims to address the limitations of traditional bankruptcy prediction models and develop a robust, automated credit reporting system tailored to SMEs. The research begins with a thorough literature review that establishes a taxonomy of datasets used in bankruptcy prediction and highlights key challenges related to data quality and integration. It then introduces an automatic feature engineering (AFE) framework to extract meaningful features from financial data, outperforming traditional financial ratio-based approaches. This is further complemented by an exploration of large language models (LLMs) for financial analysis, demonstrating their potential in calculating financial ratios, conducting Altman Z-score model and DuPont analysis, and predicting bankruptcy risk and key financial indicators with enhanced accuracy under optimized settings. Expanding beyond financial data, this dissertation integrates company adjustment behavioral data into hybrid datasets. Through uplift modeling and machine learning techniques, it reveals how non-financial factors significantly influence financial health. Considering the dynamic nature of company adjustments, MTDnet is proposed to estimate the uplift with multiple time-dependent treatments. It outperforms other uplift models establishing the necessity of considering the sequence of treatments. These findings underscore the importance of combining financial and non-financial data for comprehensive financial risk assessment. The culmination of this research is the design of an automated credit reporting system that synthesizes financial ratios, company adjustments, and predictive analytics into actionable insights. This system offers SMEs and stakeholders a scalable, data-driven tool for real-time analysis of financial health and bankruptcy risk, fostering informed decision-making and proactive risk management. By advancing methods in feature engineering, hybrid datasets, uplift modeling, and the application of LLMs, this dissertation contributes to the interdisciplinary field of data science and financial risk management. It highlights the transformative potential of integrating diverse data sources and cutting-edge technologies, paving the way for more accurate, transparent, and equitable financial systems for SMEs worldwide.