We started collecting and preparing the dataset for analysis, then we explored the information seeking for insights that might help during the model building. Next, we created a random model that predicts the rating based on the probability distribution of each rating. This model gives the worst result. We started the linear model with a very simple model which is just the mean of the observed ratings. From there, we added movie and user effects, that models the user behavior and movie distribution. With regularization we added a penalty value for the movies and users with few number of ratings. The linear model achieved the RMSE of `r RMSE(validation$rating, y_hat_mov)`, successfully passing the target of 0.8649.