Oversampling under Statistical Criteria: Example of Daily Trac Injuries Number
Julia V. Bondarenko · 2009
In imbalanced data sets, classes separated into majority (negative) and minority (positive) classes, are not approximately equally represented. That leads to impeding of accurate classication results. Well balanced data sets assume uniform distribution. The approach we present in the paper, is based on directed oversampling of minority class objects to balance non-uniform data sets, and relies upon the certain statistical criteria. The oversampling procedure is carried out for the daily trac injuries data sets. The results obtained show the improving of rare cases (positive class objects) identication with accordance to several performance measures.