Research of methods of classification of time series based on various metrics classifiers
В В Кукарцев, А С Михалев, P Yu Vaitekunaite, A I Andreev, N N Lysyannikova, P I Volegzhanin · Journal of Physics Conference Series · 2021
In this article, various approaches to the classification of time series are described and investigated in practice. This research paper discusses the operations of time series analysis using neural networks in order to improve time series, making them more accurate and fast. Also, the authors conducted a study and comparison of several metrics of various classifiers: "Method of k-nearest neighbours", "Tree solutions" and "Random forest". The following parameters were chosen as metrics for evaluating existing algorithms: "Accuracy", "Average absolute error", "F-measure". For all methods, a single dataset was selected and experiments were performed based on this single dataset, which allows an objective assessment of the work of classifiers in relation to the binary classification. We used the Python programming language with the main library SKlearn, the Ipythonnotebook application for implementing the program code as means for research in this work. It was found that there was a direct relationship between the time spent on training / testing the model and the prediction accuracy of the classifier.