A Novel Data Mining Method to Fraud Detection in Mobile Advertising

With the advent of telecommunications networks and smart phones, a new generation of digital advertising is becoming increasingly popular. Meanwhile, mobile advertising is 51% of digital market share and is expected to reach 70% by 2019. when an ad is loaded to a user s mobile device, we say an impression occurred. In recent years, a limited number of esstudihave been conducted on the detection of mobile advertising fraud, but with increase fraud methods, it s necessary to research more than before. In this study, we use a learning-based framework. Based on the raw data and the use of some derived features,labeled dataset containing fraudulent and non-fraudulent impression was created by performing the data preprocessing steps. This unbalanced dataset is then balanced by the data level balancing methods including ADASYN, in the next step, this balanced dataset with the original (unbalanced) dataset in separate experiments is given as input into SVM classifier. Finally, in this study based on evaluation criteria the best model in terms of performance among the other models is the use of ADASYN method with SVM classifier. The results show that in the final model the accuracy is equal to 71.27% and the Recall is equal to 53%. It is worth noting at the end that suggestions have been made for future work.

Keywords:

Imbalanced Data , Mobile Advertising Fraud Detection , Impression , Data Mining