The Power of Ensemble Methods: A Comparative Study of Machine Learning, Deep Learning, and LLMs for Financial Fraud Detection

This study explores the efficacy of machine learning (ML), deep learning (DL), and large language models (LLMs) for detecting financial fraud in company reports. Comparing models like a Voting Classifier ensemble, Seq2Seq, and FinBERT (an LLM tuned for financial data), we found that traditional ML ensembles outperform complex DL and LLM models in this context. The Voting Classifier achieved the highest accuracy (91.2%), followed by Seq2Seq (83%) and FinBERT (74%), underlining the strength of ensemble methods for fraud detection.