Random Walk Oversampling Technique for Minority Class Classification

Learning classifiers from imbalanced or skewed datasets is an important topic, aris- ing very often in practice in classification problems. In such problems, almost all the instances are labeled as one class, while very few instances are labeled as the other class, usually the more important class. Traditional classifiers trying to achieve an accurate performance over a full range of instances are not suitable to deal with imbalance learning tasks. They tend to classify all the data into the majority class, which is usually the less important class. Researchers have already presented many solutions to this problem both on data and algorithmic level. In this thesis a new approach to deal with imbalanced datasets is presented on the data level. This approach is an oversampling technique which involves generating new samples for the minority class by making a random walk in the dataset. The new samples are generated by some Markov Chain Monte Carlo Algorithm. Newly generated samples are then added to existing data set in order to balance the ratio between majority and minority class samples