Fitness function based on binding and recall rate for genetic inductive logic programming

Authors:
Yanjuan Li;Maozu Guo
Affiliations:
School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China,School of Information and Computer Engineering, North-East Forestry University, Harbin, China;School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China
Venue:
ICSI'12 Proceedings of the Third international conference on Advances in Swarm Intelligence - Volume Part I
Year:
2012

Citing 12
Cited 0

Learning structured concepts using genetic algorithms

ML92 Proceedings of the ninth international workshop on Machine learning
Clausal Discovery

Machine Learning - special issue on inductive logic programming
Learning Logical Definitions from Relations

Machine Learning
The relationship between Precision-Recall and ROC curves

ICML '06 Proceedings of the 23rd international conference on Machine learning
Logical and Relational Learning: From ILP to MRDM (Cognitive Technologies)

Logical and Relational Learning: From ILP to MRDM (Cognitive Technologies)
Inductive logic programming for gene regulation prediction

Machine Learning
QG/GA: a stochastic search for Progol

Machine Learning
A phenotypic genetic algorithm for inductive logic programming

Expert Systems with Applications: An International Journal
Mining associative classification rules with stock trading data - A GA-based method

Knowledge-Based Systems
Concept discovery on relational databases: New techniques for search space pruning and rule quality improvement

Knowledge-Based Systems
A comparative study on ILP-based concept discovery systems

Expert Systems with Applications: An International Journal
Improving the scalability of ILP-based multi-relational concept discovery system through parallelization

Knowledge-Based Systems

Quantified Score

Hi-index	0.00

Visualization

Abstract

The key of using genetic inductive logic programming (GILP) algorithm to learn first-order rules is how to precisely evaluate the quality of first-order rules. That is, the fitness of rules should rightly score their quality and effectively guide GILP algorithm to be close to the target rule. In this paper, a new fitness function is proposed. By adopting the concept of binding, the new fitness function can adequately utilize the information hidden in background knowledge and training examples. By considering recall rate of rules, the new fitness function can avoid generating over-specific rules. Experiments on benchmark data set show that comparing with the common fitness function based on amount of examples covered by rules, the new fitness function can measure quality of first-order rules more precisely and enhance predictive accuracy of GILP.