Deep Neural Network for Click-Through Rate Prediction

Predicted click-through rate is one of the most frequently used criteria to determine the effectiveness of an ads. In advertising production, click-through predictions are very influential for the company that places the ads. Click-through rates need to be predicted accurately because accurate prediction results determine whether the click-through rate is exactly clicked or not by the viewing consumer. Predicted click-through can be done on advertising and social network datasets. The use of these two datasets is intended to make the comparison results more convincing from the proposed method. The purpose of this study is to compare two advertising and social network datasets, by proposing the application of the Deep Neural Network (DNN) model by testing hyperparameter variations to find a better architecture in predicting whether or not users click on an advertisement. The hyperparameter variations include 3 variations of the hidden layer, 2 variations of the activation function, namely ReLuand Sigmoid, 3 variations of the optimization (RMSprop, Adam, and Adagrad),and 3 variations of the learning rate (0.1, 0.01, and 0.001). The results of experiments conducted with the advertising parameter dataset with hidden layer of 3, learning rate of 0.01,and Adam optimization with an accuracy value of 99.90%, AUC of 99.90% and Precision-Recallof99.89% while the data for social network ads parameters with hidden layer of 5, learning rate of 0.1 and Adam optimization with accuracy of 92.25%, AUC of 92.72%,andPrecision-Recallof 89.70%.