Performance Evaluation using Online Machine Learning Packages for Streaming Data

Online machine learning concept concentrates on real-time and dynamic data. This dynamic data is in the form of data streams which is a continuous flow of data of infinite length. Recently many online machine learning tools were made available for evaluating performance metrics. Most of the approaches use online machine learning packages such as MLlib, spark streaming, scikit-multiflow, Creme and River. In this paper, performance evaluation metrics are evaluated for streaming data for three popular classification models: Logistic Regression (LR), K-Nearest Neighbors (KNN) and Gaussian Naive Bayes. For the evaluations, Creme and River online machine learning packages have been used. In the implementation phase, metrics are evaluated for four popular datasets: Elec2, Phishing, Bananas, and CreditCard. The accuracy, ROC-AUC, precision, recall and F1-score performance metrics are used for the comparison of the online classifiers. After analyzing the results of the four datasets, it is observed that the performance evaluation using the River online machine learning package is more accurate and reliable.