IMPROVING DATA MINING PERFORMANCE FOR HIGH UTILITY ITEM SETS USING NOVEL DEEP LEARNING FRAMEWORK

Abstract

Traditional data mining approaches mainly concern with the frequent pattern’s extraction, but these approaches fail to take into account the importance of utility of items. The contribution of this study is a novel deep learning algorithm to enhance the high utility item set mining performance. An effective capturing the relationship between items and its utilities is a undertaken by the proposed algorithm using deep learning approach. The hybrid algorithm significantly enhances the accuracy and efficiency of high utility item set mining by combining convolutional and recurrent neural networks. Deep learning achieves this with the automatic feature extraction and hierarchical learning. The proposed model obtained 87.69% accuracy, 88.69% precision, 87.35% recall and 87.40% F1-Score. This section describes empirical results using several datasets proving the effectiveness of the proposal, reporting achieving better value than baseline methods, both in terms of utility and run time, in terms of design by stages. This algorithm can significantly optimize the data mining of high utility item sets bringing a major success to real-world applications as well.