Malayalam Fake Review Detection Using Hybrid Feature Engineering and Stacked Ensemble Learning

Abstract

The rapid growth of e-commerce platforms has significantly increased the dependence on customer reviews, making the detection of fake reviews essential for preserving consumer trust and ensuring the credibility of online marketplaces.While substantial progress has been made for high-resource languages, limited progress has been made for low-resource and morphologically complex languages such as Malayalam. This paper presents a hybrid feature engineering and stacked ensemble learning framework for Malayalam fake review detection. A Malayalam dataset was constructed by translating a labeled subset of the Amazon Deception Dataset to enable experimental evaluation. A stacked ensemble architecture comprising of Logistic Regression, Support Vector Machine, Random Forest, and Extra Trees, with XGBoost as the meta-classifier, was employed for classification. Experimental results achieved a test accuracy of 94.0%, while cross-validation confirmed stable generalization performance. The findings demonstrate that hybrid feature engineering combined with stacked ensemble learning provides an effective and practical framework for fake review detection in low-resource language settings.