Fake Mobile Product Review Detection Using Machine Learning

Online product reviews are an important source of information for consumers before making purchasing decisions. However, the growing availability of misleading, promotional, spam-like, and otherwise unreliable reviews can reduce consumer trust. This study presents a lightweight machine-learning approach to classify mobile-phone product reviews as Genuine or Fake, based on the experimental labels used in the dataset. Review text was cleaned and represented using term frequency–inverse document frequency (TF–IDF). Naive Bayes and Logistic Regression were compared using accuracy, precision, recall, and F1-score. Evaluation of two feature configurations: text-only features and text combined with eight mobile-specific parameters—battery, camera, display, performance, heating, charging, storage, and price. Logistic Regression outperformed Naive Bayes in the reported comparison. Its text-only configuration achieved 92.225% accuracy, while adding the mobile-specific parameters increased accuracy to 92.450% and improved the fake-class F1-score from 0.817595 to 0.825231. The final model was integrated into a simple web application that returns a predicted class and confidence value. Because the dataset labels were generated through the project’s experimental labelling procedure rather than independently verified reviewer investigations, the results measure agreement with those labels and should not be interpreted as proof that a review is fraudulent in the real world.