OPECUR: An Enhanced Clustering-Based Model for Discovering Unexpected Rules

• Published in: Advanced Data Mining and Applications Vol. 13088; pp. 29 - 41
• Main Authors: Darrab, Sadeq, Bhardwaj, Priyamvada, Broneske, David, Saake, Gunter
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Lecture Notes in Computer Science
• ISBN: 9783030954079; 3030954072
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-95408-6_3

Rare association rules gained importance since they are widely used in critical real-life domains, such as medicine, fraud detection, malware attacks, recommender systems, and weather forecasting. In order to extract actionable rules that can be used in real-life scenarios, user confidence and an easy-to-use model with as few as possible tuning knobs are required. On top of this, an inherent imbalance of the data (e.g., in the medical domain, there are fewer ill people compared to healthy people) poses a severe challenge, which complicates the finding of rare patterns. Recently, an unsupervised clustering model was proposed to discover interesting rare rules. However, the performance of this model degrades in terms of time and accuracy. In this paper, we propose an efficient model to recover interesting rare rules. In this model, we employ machine learning-based classifiers to assess the performance of the generated rules. To evaluate the proposed model, we experiment with three real-life medical datasets. The experimental results show that our model outperforms the state-of-the-art model in terms of time and accuracy. Furthermore, we generate more accurate results, which means that the user is only confronted with the most important and compact rules. This reduces a user’s effort on postprocessing of associating rules significantly.