Improving SVM Classification on Imbalanced Data Sets in Distance Spaces

Abstract

Abstract-Imbalanced data sets present a particular challenge to the data mining community. Often, it is the rare event that is of interest and the cost of misclassifying the rare event is higher than misclassifying the usual event. When the data is highly skewed toward the usual, it can be very difficult for a learning system to accurately detect the rare event. There have been many approaches in recent years for handling imbalanced data sets, from under-sampling the majority class to adding synthetic points to the minority class in feature space. Distances between time series are known to be non-Euclidean and nonmetric, since comparing time series requires warping in time. This fact makes it impossible to apply standard methods like SMOTE to insert synthetic data points in feature spaces. We present an innovative approach that augments the minority class by adding synthetic points in distance spaces. We then use Support Vector Machines for classification. Our experimental results on standard time series show that our synthetic points significantly improve the classification rate of the rare events, and in many cases also improves the overall accuracy of SVM.

An Empirical Analysis of Imbalanced Data Classification

“…This is due to the imbalanced numbers of instances. On the contrary, G-means and F-measure have been adopted as the experimental metric for class imbalance (Koknar-Tezel Latecki, 2009;Imam, et al,2006). Moreover, which metric should be applied to the experiment depends also on the classification purpose and the domain of the benchmark datasets.…”

“…1; the minority class to build a less imbalanced data (Akbani, et al, 2004;Haibo Garcia, 2009;Koknar-Tezel Latecki, 2009). In addition, sampling methods have been proved to improve the classifier performance (Haibo Garcia, 2009;Koknar-Tezel Latecki, 2009). For example, in (Japkowicz, 2000), the author examined the dual techniques of under-sampling and over-sampling and the empirical analysis demonstrated the efficiency of both methods.…”

“…Fraudulent behaviours, such as shill bidding and bidder collusion, usually result in a negative impact on honest users such as money loss and waisted effort (Mamum Sadaoui, 2013). Learning from training data that are imbalanced is diffcult since the standard machine learning systems often misclassify minority instances as majority ones (Koknar-Tezel Latecki, 2009). This means that the prediction of classifying a new data into the minority class is very low (Haibo Garcia, 2009).…”

“…In many real-life situations, the cost of incorrectly classifying the minority event is much higher than misclassifying the normal event, for example misclassifying a fraudulent bidder as an honest one. As a result, many researchers developed various techniques to correctly classify imbalanced datasets, such as data sampling methods that have been proven to improve the classifier performance (Koknar-Tezel Latecki, 2009;Haibo Garcia, 2009). On the other hand, the imbalance problem is not the only factor that impacts the classification accuracy, but the data complexity plays a more significant role (Haibo Garcia, 2009).…”

“…Interesting research works have been carried out in order to increase the learning rate of the minority events by extending and adapting the SVM classifier (Koknar-Tezel Latecki, 2009;Imam, et al,2006;Haibo Garcia, 2009). In online auctions, we are interested in classifying fraudulent users who represent the minority category.…”