Data distributions automatic identification based on SOM and support vector machines

Abstract

It is very important to identify probability distributions fast and efficiently in data analysis. The paper analyzes data distributions automatic identification using a combined structure mode via self-organizing map and support vector machines. First, the paper sets up data distributions identification training sets, which are based on summary statistics including kurtosis, skewness, quantile and cumulative probability. Then, different data distributions are clustered using a self-organizing map. Furthermore, the clusters are learned and classified respectively using support vector machines. Finally, identification of random data distribution time series is tested in combined structure mode. The results indicate that the approach of the paper is feasible and efficient for automatically identifying data distributions in comparison with other methods.