A NOVEL META-LEARNING SYSTEM FOR CLUSTERING ALGORITHM RECOMMENDATION BASED ON META-FEATURES

Abstract

Clustering a novel unsupervised problem usually necessitates prior knowledge of a factor that has a significant impact on the results achieved. Since clustering is one of the most special areas where Machine Learning (ML) algorithms are used, it is very important to understand high quality clustering algorithms. Trial and Blunder can be used to determine high-quality rules for modern datasets by experimenting with large algorithms with some bias, but this approach comes with high computational costs. With different biases, however this method has a significant computing cost. A unique meta-learning system based on meta-features is described in this research for clustering algorithm recommendation. This method may use meta-features to rank algorithms based on their suitability for a new dataset, and the resulting ranking can be used to successfully select the best ML algorithm for calculating the number of clusters in unsupervised data. To review and analyze datasets and extract a set of meta-features, this system employs a meta-learning method. These meta-features can then be used to intelligently select an effective clustering approach, eliminating the need to manually execute the method. The experiment can be carried out on datasets to identify expected results based solely on their meta-features for clustering algorithm selection utilizing meta-learning. After then the accuracy and runtime of the results are ranked individually