Weighting and Pruning of Decision Rules by Attributes and Attribute Rankings

Abstract

Abstract. Pruning is a popular post-processing mechanism used in search for optimal solutions when there is insufficient domain knowledge to either limit learning data or govern induction in order to infer only the most interesting or important decision rules. Filtering of generated rules can be driven by various parameters, for example explicit rule characteristics. The paper presents research on pruning rule sets by two approaches involving attribute rankings, the first relaying on selection of rules referring to the highest ranking attributes, which is compared to weighting of rules by calculated quality measures dependent on weights coming from attribute rankings that results in rule ranking.

Pruning Decision Rules by Reduct-Based Weighting and Ranking of Features

“…Still other, more complex quality measures can be defined for rules through the attributes included in their premises, their properties, or combinations thereof [ 49 ]. Scores assigned to individual rules can lead to their weighting, and then, rule ranking can be exploited to discard less important rules and keep only some subset of the most advantageous elements.…”

Calculating the relative importance of condition attributes based on the characteristics of decision rules and attribute reducts: Application to crowdfunding

“…In most of the existing literature, rule attractiveness measures have been investigated from the perspective of rules pruning, especially when there is a large number of decision rules, in order to eliminate irrelevant, weak or obvious ones; see for example [108]. A second application of rules attractiveness measures is feature extraction, where the characteristics of decision rules are used to identify the most important features; see for example, [70] [95].…”

“…In this paper, rules attractiveness measures are used to induce importance of condition attributes. At this level, it is important to note the existence of some works that use attributes rankings as in [107] [108], or the importance of elementary conditions as in [99] as basis for rule selection or redefinition.…”

Weight empowerment method in information fusion for radar‐seeker performance evaluation

“…However, measuring the effectiveness of a radar seeker using only its theoretical performance obtained in the laboratory test is not sufficient. The index system used in the present work has been perfected and supplemented according to the existing literature to make it more suitable for performance evaluation of a radar seeker [8].…”