Evaluation of Hybrid Recommendation System and Machine Learning Algorithms for E-Commerce Platform

E-commerce industry has become more popular over the last 20 years with the increased popularity of the Internet. Thus, the Internet facilitated retail for individuals who are able to order products online that get delivered to their own home. This made users’ experience crucial in e-commerce industry. In order to understand users’ experience and improve it, e-commerce industry has been using business tools such as recommender systems. These systems have a challenge when applied on their own to big datasets. Consequently, this research combined the use of a hybrid recommender system and machine learning analytics for predicting users’ preferences for products and products’ popularity from an e-commerce dataset. It is worth noting that the hybrid recommender system combined both collaborative and content-based filtering. In this respect, both random forest and alternative least square models were applied to LightFM and Sparks recommender systems, respectively. Hyperparameter tuning was crucial for optimizing random forest model that gave an AUC in the range of 0.6–0.96. In addition, alternative least square model was evaluated by the root mean square error score that was 0.259. Hence, random forest model applied to hybrid recommender system outperformed the alternative least square-based model. The findings showed to be useful for large e-commerce datasets where the hybrid model showed that it can handle large volumes of datasets without affecting performance accuracy.