Sentiment analysis is a key area of natural language processing that enables the classification and extraction of opinions from text. This paper presents a comprehensive empirical study comparing the performance of classical machine learning models and advanced neural network architectures on the IMDB dataset. The experimentation covers multiple approaches, including Logistic Regression, Naive Bayes, and Random Forest for classical models, as well as CNN and LSTM for neural architectures. The models are evaluated using various metrics (accuracy, precision, recall, F1-score, and training time), highlighting trade-offs between speed, accuracy, and computational complexity. The results show that while neural models capture complex linguistic relationships more effectively, classical models remain competitive for tasks requiring fast and efficient execution. This study provides practical guidelines for selecting models based on application constraints and explores potential improvements, including the integration of hybrid models and the optimization of neural architectures.