This study examines the impact of social media sentiment on stock prices in the banking and finance sector of Turkey, focusing on stocks that have been detected to have been manipulated by the Capital Markets Board (CMB). Using comments from Twitter and Investing.com, the study analyzes both general sentiment trends and specific suspicious actions associated with stocks. Using a binary sentiment classification approach, the study distinguishes between (1) three classification models (positive, neutral, negative) and (2) a binary classification that identifies suspicious and neutral comments. Advanced machine learning and deep learning models are used in the study to generate predictive insights. Testing various feature combinations, the study finds that the most successful models are LSTM and CNN, achieving the highest predictive accuracy when Twitter sentiment is combined with the suspicious score, and outperforming traditional models based solely on historical price trends and Investing.com sentiment data. The results demonstrate the strong predictive potential of Twitter and highlight the role of social media in indicating and potentially driving suspicious market behavior. This study highlights the need to monitor sentiment trends in markets, providing practical insights for investors and regulators.