Statistical debugging using a hierarchical model of correlated predicates

  • Authors:

  • Saeed Parsa;Maryam Asadi-Aghbolaghi;Mojtaba Vahidi-Asl

  • Affiliations:

  • Department of Software Engineering, Iran University of Science and Technology, Tehran, Iran;Department of Software Engineering, Iran University of Science and Technology, Tehran, Iran;Department of Software Engineering, Iran University of Science and Technology, Tehran, Iran

  • Venue:
  • AICI'11 Proceedings of the Third international conference on Artificial intelligence and computational intelligence - Volume Part I
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

The aim of statistical debugging is to identify faulty predicates that have strong effect on program failure. In this paper predicates are fitted into a linear regression model to consider the vertical effect of predicates on each other and on program termination status. Prior approaches have merely considered predicates in isolation. The proposed approach in this paper is a twostep procedure which includes hierarchical clustering and the Lasso regression method. Hierarchical clustering builds a tree structure of correlated predicates. The Lasso method is applied on the clusters in some specified levels of the tree. This makes the method scalable in terms of the size of a program. Unlike other statistical methods which do not provide any context of the failure, the predicates contained in the group that is provided by this method can be used as the bug signature. The method has been evaluated on two well-known test suites, Space and Siemens. The experimental results reveal the accuracy and precision of the approach comparing with similar techniques.