This project investigates whether machine learning can predict whether a newly published book will become a bestseller using only information available at or near publication, before sales or reader feedback exist. We merged the GoodBooks dataset with the NYT Bestseller archive to build a curated dataset of 7,983 fiction books. After comparing logistic regression, Random Forest, and Gradient Boosting models, Random Forest achieved the best overall performance, substantially outperforming a naive baseline. The most predictive features include an author’s prior bestseller history, page count, series membership, and genre. Our findings demonstrate how machine learning can support publishing decisions such as manuscript acquisition, marketing investment, and inventory planning while highlighting opportunities for future improvements in data coverage and model development.