Relationship between structural complexity and performance of data mining classification algorithms

Abstract

Data mining algorithms have been applied in industries, government, military, retail, banking and educational contexts for analysis, prediction and classification of data. Little or no attention has been given to the structural complexity of those algorithms and the effect it has on their performances. Hence, this study investigates the relationship between structural complexity and classification performance of Multi-Layered Perceptron, J48 decision tree, Naive Bayes and REPTree data mining classification algorithms. Their performances were analysed based on training time and percentage accuracy, using Nursery and Glass datasets from online UCI data repository. Structural complexities of these four data mining classification algorithms were computed using Java programming language. Relationships between structural complexities of the classification algorithms and their training times as well as accuracy of classification/prediction were obtained. Briefly, the results showed that the higher the structural complexity, the higher the training times and accuracies of the algorithms.

Keywords: Big data, Data mining, Structural Complexity, Classification algorithms