Step-Factor Resampling Technique for Imbalanced Sequence Data Classification

Abstract

The mobile ad fraud detection plays a vital role in research community nowadays since a large sum of money is being circulated in this industry. The fraudsters who generate illegal revenue by executing various types of fraudulent activities are difficult to be detected. Various researches are being conducted by academics and industry experts to find optimum solutions to tackle these fraudulent activities. One of the barriers they face is the class imbalance problem in the datasets. In an imbalance dataset, the frequency of the states of the target class varies significantly. A particular dataset which is suffered with class imbalance problem will mislead the final results of the predictive model. The researchers have proposed a number of methods to address this class imbalance problem. A novel technique which is composed with smoothing and resampling techniques has been proposed in this study to solve the class imbalance problem in the context of sequence data. The proposed technique is evaluated with a hidden Markov scoring model—HMSM. The model shows significant improvement in accuracy, recall, precision and F-score with this novel resampling technique and smoothing approach. Moreover, the model becomes more sensitive towards the specificity which is the key factor of any kind of fraud detection methods.