Progress Report: Predicting Which Recommended Content Users Click

Machine learning models can be used to predict which recommended content users will click on a given website. The given dataset contains millions of samples that map some feature about an ad or web page to a number. We reduced this dataset to a more manageable size to minimize computation time, and we extract features based on this reduced set. The features we extracted are based on the advertisers and campaigns associated with the advertisements in the dataset. We initially built models based on Naive Bayes and logistic regression. We also built a model based on the support vector machine (SVM) using hinge loss. In addition, we constructed a neural network using the multilayer perceptron model to capture the non-linearity of features in order to obtain a better prediction score. The best result we obtained was using SVMs, and it yielded an accuracy of 0.46. The result can be scaled up to the complete dataset by optimizing our implementation using parallel computing.