TCNAttention-Rag: Stock Prediction and Fraud Detection Framework Based on Financial Report Analysis

Due to the high volatility of financial markets and the prevalence of financial fraud, real-time stock market forecasting for listed companies remains a challenging task. To address these challenges, this study proposes TCNAttention-RAG, a hybrid deep learning framework integrating Temporal Convolutional Network (TCN), Multi-Layer Perceptron (MLP), Attention Mechanism, and Retrieval-Augmented Generation (RAG) for enhanced stock price forecasting. The model leverages TCN for temporal feature extraction, MLP for nonlinear representation, and Attention for feature weighting, while RAG dynamically retrieves key financial insights from corporate reports to improve predictive accuracy. Using NASDAQ-listed stock price data (2014–2020), combined with corporate financial reports, market transaction data, and macroeconomic indicators, a multi-dimensional dataset is constructed. Experimental results demonstrate that TCNAttention-RAG outperforms traditional models in accuracy and recall, effectively capturing stock price fluctuations. Despite its limitations in handling extreme market events, the model exhibits high reliability and predictive robustness. This study introduces a multi-modal data-driven approach to financial forecasting, offering insights into intelligent financial analysis and enhancing decision-making in volatile markets.