Evaluating Machine Learning Algorithms for Time Series Classification Using Complex Network Features

Time series classification is a crucial step in many scientific and engineering fields because it allows us to identify and predict trends or patterns of behavior in data over time. This study aims to explore the efficacy of five popular machine learning algorithms - Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Linear Regression (LR), Naive Bayes (NB), and Decision Tree (DT) - in the task of time series classification, with a new approach to feature extraction based on complex networks. The research method involves applying feature extraction techniques using complex networks to graph and compare the performance of these algorithms on prepared datasets. The results show significant differences in classification efficacy among the algorithms, with the highest accuracy in this time series classification study reaching 96.92%. The study concludes that selecting the appropriate algorithm and feature extraction method can significantly improve the quality of time series classification, paving the way for more effective future developments in time series processing models.