In the era of information revolution that we live in, data and information are constantly increasing at every moment to the point that data and information have become overloaded. It is difficult for the users to access the items they like and search for on the Internet. Therefore, recommendation systems are an indispensable tool to help the users and suggest items that are likely to be their favorite. Hence, giant technology companies such as Netflix, YouTube, and Amazon Prime rely heavily on such systems to recommend video content according to the tastes of their users in order to increase their revenues. One of the common types of recommendation systems is collaborative filtering, which addresses the problem of information overload and relies on calculating the similarity between the active user and other users. Similarity measures are important components that have a great impact on the accuracy of the suggestions resulting from these tools. Similarity measures are also responsible for finding the degree of similarity between users or items. Therefore, we must choose an appropriate similarity measure, in order to improve the accuracy of collaborative filtering. This paper proposed and implemented an adaptive predictive KNN approach which based on item based Collaborative filtering for improving the rating score prediction accuracy. This approach uses different three common similarity measures such as Cosine, Pearson Correlation Coefficient (PCC), and Mean Squared Difference (MSD). The objective of this research is to analyze and compare the proposed approach results. The accuracy is evaluated by performing experiments on popular benchmark dataset. The experimental results show that the proposed approach achieves better accuracy when using the similarity measures Cosine, Pearson Correlation Coefficient (PCC), and Mean Squared Difference (MSD) by 0.937, 0.934 and 0.928 respectively in terms of RMSE and by 0.736, 0.732 and 0.728 respectively in terms of MAE.