Sentiment analysis is crucial for interpreting textual data, with applications across domains. The proposed work investigates various sentiment analysis methodologies, spanning from traditional rule-based approaches to advanced deep learning techniques. Using the sentiment140 dataset, the proposed work compares traditional TF-IDF with fine-tuned DistilBERT. The work details DistilBERT’s architecture for sequence classification. Evaluating the models using accuracy, precision, recall, and F1 score reveals promising performance. Logistic regression (LR) achieves 82% accuracy, with true positives (TP) at 41.9%, false positives (FP) at 9.05%, and false negatives (FN) at 8.11%. Similarly, Linear Support Vector Classification (Linear SVC) achieves 83% accuracy with true positives (TP) at 41.37%, false positives (FP) at 9.51%, and false negatives (FN) at 8.64%. In contrast, DistilBERT achieves an accuracy of 86%. Its superiority lies in determining sentence sentiment probabilities, supported by its vocabulary range and precision. This study highlights the importance of additional layers in model performance and elucidates sentence-level sentiment probabilities. By showcasing deep learning’s strength in handling nuanced emotions, this research advances sentiment analysis methodologies.