An Effective Feature Selection and Data-Stream Classification Model HDP

Abstract

Especially with the development of the Internet, the data source generated from the Internet becomes much huger than before. Because of the infinite number and difficult storage of samples, the routine classification algorithms are unable to handle such amount of data samples. Additionally, the complex data source’s features also make the data process inefficient. In this paper, we propose a new model HDP to solve the difficulties mentioned above. In HDP model, the HSF feature selection algorithm is responsible for reducing the number of the samples' features, and improved Hoeffding-ID classification algorithm focuses on the data classification progress. Experiment results show that HDP model raises the classification accuracy, additionally the time cost of this model also gets a better performance. Besides that the memory usage of the model is not increasing with the number of data samples. All above demonstrate that HDP model is qualified as the data-stream classification solution.