Facets of Fakes in Cyberspace: Machine and Ensemble Learning-Based Decisions and Detections

Abstract

Fake online reviews hinder internet marketing efforts to build businesses and brands in a competitive market with changing consumer expectations. This helps brands attract clients, making fake online reviews hard to uncover. Hence, fake reviews and websites are extensively examined. AI models like the Generalized Additive2 Model (GA2M) and its ensemble with the Elastic-net Classifier model have been studied using Log-Loss metric. This research, analysis, and depiction help demarcate bogus hotel reviews and websites from genuine entities. The paper uses ML classifiers (Decision Tree, Logistic Regression, Naïve Bayes) and ensemble models (Random Forest, Gradient Boosting) to identify legitimate websites using binary classification. This article compares ML classifiers and ensemble models by accuracy, precision, recall, f1-score, and ROC-AUC to evaluate their pros and downsides. Elastic-Net Classifier (L2 / Binomial Deviance) with score of 0.2879 outperformed GA2M model by 0.66% in LogLoss holdout score on Hotel dataset. LogLoss predicts values better than ROC-AUC due to its closer proximity to predicting actual values. Elastic-Net Classifier (L2 / Binomial Deviance) surpassed GA2M in F1 score, precision, and accuracy by 0.4%, 1.84%, and 0.63%. Ensemble techniques outperform ML classifiers in the Fraudulent and Legitimate Online Shops dataset with ROC-AUC scores of 0.71%, 1.73%, 0.76%, 1.10%, and 0.63% using 50% to 90% training datasets and 50% to 10% holdout datasets.