An Enhanced SMOTE Algorithm Using Entropy and Clustering for Imbalanced Accident Data

Over the course of the century, many real-world applications of imbalanced data are emerged. One of its implication which is first considered in this context, is imbalanced accident data. In this paper, the data of transportation and accidents in Tehran-Bazargan highway between 2010 and 2015 is considered. In the pre-processing step, SMOTE is considered as one of the most important over-sampling technique that effectively balance the imbalanced data. However, it brings noise and other problems and a great need is felt for improving this method. To solve these problems, several techniques have been proposed in this study such as combination of dynamic selected, weighted attribute and distance weighted techniques along with mixture of classification and clustering techniques. Performance of the proposed algorithm is measured by f-measure and ROC curve and the results are compared by Weka’s SMOTE with different algorithms.

Keywords:

imbalanced data , SMOTE , f-measure , ROC curve