Comparative Analysis of Machine Learning and Deep Learning Models for Financial Sentiment Classification

Financial sentiment analysis plays a vital role in understanding market behavior, investor psychology, and decision-making in the digital economy. This study presents a comparative evaluation of Machine Learning (ML) and Deep Learning (DL) models for financial sentiment classification using a dataset of 5,842 financial sentences labeled as positive, negative, or neutral. The objective is to assess the performance of traditional ML algorithms and a DL architecture in predicting sentiment polarity within financial texts. Four models were implemented, including Logistic Regression, Support Vector Machine, and Random Forest as ML models, and Long Short-Term Memory (LSTM) as the DL model. The ML models utilized TF–IDF feature extraction, while the LSTM model employed word embeddings to capture semantic relationships. Model performance was measured using accuracy, precision, recall, and F1-score. The results show that Logistic Regression achieved the highest accuracy at 67.41 percent, followed by SVM with 65.95 percent, Random Forest with 59.54 percent, and LSTM with 53.55 percent. Confusion matrix analysis revealed that all models performed best on neutral sentiments but struggled to identify negative ones due to dataset imbalance. These findings suggest that traditional ML models remain effective and computationally efficient for financial sentiment analysis, particularly when data is limited or unbalanced. Although DL models offer theoretical advantages in capturing contextual dependencies, their performance depends heavily on dataset size and quality. This research concludes that Logistic Regression provides a strong and interpretable baseline for sentiment classification, while future studies should explore transformer-based architectures to enhance contextual understanding and predictive accuracy.