In the realm of e-commerce, ensuring recommendation integrity and enhancing user experience are vital considerations. This paper introduces a novel approach to product recommendation leveraging unsupervised learning techniques applied to the Flipkart Mobiles Dataset. Unlike traditional outlier detection, our focus shifts to the realm of product selection based on customer reviews. The proposed framework utilizes a blend of unsupervised algorithms, including Autoencoders, DBSCAN, Isolation Forest, One-Class SVM, Local Outlier Factor, and Logistic Regression, to analyze and recommend products without the need for labeled data. Autoencoders extract meaningful representations from the review data, capturing subtle patterns and nuances in customer feedback. DBSCAN groups similar products based on review characteristics, effectively identifying clusters within the dataset. Isolation Forest excels at isolating distinct products, particularly in high-dimensional review spaces. One-Class SVM constructs a model of typical review patterns, flagging deviations as potential recommendation candidates. Local Outlier Factor measures local anomalies in review data, aiding in the identification of noteworthy products based on variations in review density. Logistic Regression complements these unsupervised techniques by estimating the likelihood of a product being recommended based on its features. This framework aims to optimize product recommendations on e-commerce platforms, enriching user engagement and satisfaction. Experimental validation using the Flipkart Mobiles Dataset showcases the efficacy of the approach in delivering personalized and relevant product suggestions to users, thus fostering trust and loyalty in the e-commerce ecosystem.

E-commerce, Predictive Analytics, Transactional Patterns, Dynamic Outlier Detection, Product Recommendation.