PRESENTING IMPROVED A COMBINED STATISTICAL/LEARNING APPROACH TO DETECT FAKE FEEDBACK IN CLOUD ENVIRONMENT USING HIERARCHICAL CLUSTERING WITH RSA AND EXTREME LEARNING MACHINES WITH PSO

Abstract

The increase in fake feedback in cloud environments is one of the main challenges in maintaining user trust and ensuring service quality. Therefore, developing intelligent approaches to distinguish real feedback from fake has become increasingly important. In this study, a combined framework based on statistical analysis and machine learning is presented for fake feedback detection, in which the improved hierarchical clustering with reptile search algorithm (AHC-RSA), the E-EDA statistical model for anomaly filtering, and the optimized extreme learning model with particle swarm optimization (ELM-PSO) are integrated. This combination has reduced intra-cluster dispersion, improved data separation, increased prediction accuracy, and accelerated model convergence. Evaluations on two datasets, CloudArmor and Epinions, showed that the proposed model achieved 99.83% and 99.84% accuracy, respectively. The results show that the model is superior to SVM, ANN, LSTM and GRU algorithms in terms of accuracy, stability and convergence speed. On average, the proposed model has about 5 to 8 percent improvement in accuracy and F1-score and a significant reduction in MSE and RMSE errors compared to the best previous methods. As a result, the proposed framework can be used as an effective and reliable method for detecting fake feedback and improving trust in cloud services.