This paper explores machine learning models for sentiment analysis of Amazon book reviews, addressing challenges related to mixed and ambiguous sentiments in unstructured text. Sentiment analysis provides valuable insights from customer feedback that influence purchasing decisions and product development. Preprocessing steps, including HTML removal, tokenization, and non-alphabetic character elimination, were applied to prepare a dataset of 100,000 Amazon book reviews. A Random Forest classifier was used to categorize reviews into positive and negative sentiments. A key contribution is the introduction of a category-specific approach to explore sentiment variations across genres and authors. An optimized preprocessing pipeline was developed to reduce noise while preserving critical information. The model achieved an accuracy of 83.26% and a high F1-score of 90% for positive sentiment, though recall for negative sentiment was lower at 39%. This discrepancy is attributed to challenges such as class imbalance and subtle expressions of negative sentiment. Further analysis revealed that positive reviews were more common in fiction, and some authors received polarized feedback. These findings highlight the potential of machine learning in improving sentiment analysis, aiding consumer decision-making, and assisting sellers and publishers in refining strategies and addressing recurring issues.