The Impact of Rule Evaluation Metrics as a Conflict Resolution Strategy

• Published in: New Trends in Information and Communications Technology Applications Vol. 1183; pp. 105 - 127
• Main Authors: Al-A’araji, Nabeel H., Al-Mamory, Safaa O., Al-Shakarchi, Ali H.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2020; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Classification; Rule evaluation metrics; Rule extraction; Rules conflict resolution
• ISBN: 3030553396; 9783030553395
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-030-55340-1_8

It is crucial for each rule induced via machine learning algorithm is to be associated with a numerical value(s) which can reflect its properties like accuracy, coverage, likelihood. The accumulation of these properties is the so-called evaluation metrics. These metrics are important for both rule induction systems (for stopping rules generation) and rule classification systems (for solving rules conflict). This paper describes the most important of both statistical and empirical rule evaluation metrics. Thereafter, the paper presents an approach that utilizes and shows the impact of these metrics as a rule conflict resolution strategy during classification tasks when combining two heterogeneous classifiers (Naïve Bayes (henceforth, NB) and decision tree J48). To accomplish this goal, authors have extracted rule-set from J48 and presented a new method for rule construction from NB. Experiments have been conducted on (WBC, Vote, and Diabetes) datasets. The experimental results show that different evaluation metrics have a different impact on the classification accuracy when used as a conflict resolution strategy.