Sentiment analysis plays a crucial role in automatically categorizing user sentiments, such as positive, negative, or neutral, towards various subjects, products, or newsworthy items. Machine learning, an effective artificial intelligence (AI) technique, has been widely adopted to meet the increasing demand for accurate emotional analysis. Analyzing user sentiments on social networks like Facebook and Twitter have proven to be a dominant tool for gaining insights into public opinion and has applications across various sectors. However, challenges in natural language processing hinder the precision and effectiveness of emotional analysis. In this study, a dataset of 50,000 reviews was scraped from the IMDb website. VADER, a widely used sentiment analysis tool, was employed to label the data as positive or negative based on polarity. Preprocessing techniques were then applied, including the removal of URLs, spell correction, punctuation, single character and double space removal, digit removal, tokenization, stopword removal, and lemmatization. For feature extraction, an LSTM model was utilized due to its ability to capture sequential information effectively. The extracted features were passed to a voting classifier, consisting of XGBoost and LightGBM models, for classification. The hybrid model reached a remarkable accuracy of 94% on the testing data. Performance metrics were used to evaluate the performance of the hybrid model. The precision, recall, and F1 score on the testing data were reported as 0.93997, 0.9397, and 0.93983, respectively, indicating the high quality of sentiment classification.