An Optimization Method Based on Drift Data and Time Series Information

With the rapid development of Internet technology, a large amount of streaming data with potential application value will be generated on the Internet. Compared with traditional static data, streaming data typically exhibits the phenomenon of concept drift. This makes it challenging for traditional machine learning methods to uncover the potential application value of streaming data. To solve this problem, previous literature has proposed various methods, such as drift detection, ensemble learning, and sliding window methods, to address the issue of concept drift. However, most existing methods have ignored optimizing the initial data, resulting in the model being unable to accurately describe the distribution of current time data. This leads to a decrease in the model’s accuracy. In this paper, we propose a novel optimization method called Optimization Method Based on Drift Data and Time Series Information (OMDD-TSI) to overcome the shortcomings of previous algorithms. Our optimization method is divided into two steps. First, determine the position of the drifting data in the initial data according to the Hoeffding’s inequality, and supplement the drifting data based on the classification uncertainty. Secondly, our method assigns weights to drift data based on time series information. The results of a simulation experiment on 12 synthetic stream data and 5 real-world stream data show that the accuracy of the previous state-of-the-art methods has been improved to varying degrees after being optimized by our method (OMDD-TSI).