Based on the preferences and previous history with many other customers, recommendation system assists users in quickly finding pertinent goods. They benefit users by making searching easier and companies by encouraging the sale of their products. Various filtering methods, such as collaborative, demographic, and content-based filtering are used to create a prediction model. Furthermore, because of concerns with variance and sparsity, massive amounts of information may result in suggestions with a limited degree of accuracy. In this work, we analyze random forest (RF) and AdaBoost classifiers and suggest a hybrid learning technique that uses user-to-user filtering features to identify the closest users. This technique is designed as a result of research into how to lower rating prediction mistakes based on user prior interactions, thus increasing prediction performance in two classification algorithms. We used a characteristic combination technique to increase predictability. Our proposed model has attained less error rate on using random forest and AdaBoost classifiers for both mean absolute error and root mean squared error filters.