Sentiment analysis, a critical component of natural language processing, holds significant relevance across diverse domains. In the digital age, the ability to decipher emotions encoded in textual content has become a valuable asset. This study explores sentiment analysis as a fundamental aspect of natural language processing (NLP), employing machine learning techniques to reveal sentiments ranging from positivity and negativity to more nuanced emotional expressions. Three distinct approaches—Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and the revolutionary BERT model—are employed for in-depth investigation. The study utilizes a carefully curated subset of the Amazon reviews dataset, encompassing over 50,000 rows with star-based ratings ranging from 1 to 5. Acknowledging the inherent imbalance in the dataset, characterized by a predominant number of positive reviews, the research aims to evaluate the effectiveness of each model in capturing nuanced sentiments.