With the continuous evolution of E-commerce systems, online reviews play a vital role in shaping consumer decisions and maintaining a business's digital reputation. Positive reviews often attract more customers, significantly boosting sales. However, the rise of deceptive or fake reviews often crafted to manipulate perception and build a false reputation poses a serious challenge. Detecting such reviews has become a critical and active area of research. While earlier approaches focused on traditional machine learning models, this paper advances the field by incorporating both classical and deep learning techniques. In addition to feature extraction and reviewer behavior analysis, we evaluate the performance of machine learning models such as K-Nearest Neighbors (KNN), Naive Bayes (NB), and Logistic Regression. Furthermore, we integrate cutting-edge transformer-based models including GPT-2, RoBERTa, XLNet, and DistilBERT for deeper contextual understanding. Experimental results on a real Yelp dataset demonstrate that transformer models significantly outperform classical models in detecting fake reviews with higher accuracy and robustness.

KEYWORDS: Machine Learning, Fake Reviews, Deep Learning, GPT-2, RoBERTa, XLNet, DistilBERT, Logistic Regression, NLP.