Attribute rank-based weighted decision tree

Abstract

Decision tree is a renowned classifier reformed by innumerable tree generation techniques to construct efficient trees. In this paper, a novel framework for constructing decision tree classifiers using ranked attributes and weights is presented. The tree is fabricated using proposed framework hierarchically from root to leaf using ranks and weights of attributes assigned to branches based on their contribution towards classification accuracy. The weights are perpetually updated until the tree furnishes maximum classification accuracy. This framework is experimented with cuckoo search, firefly search, wolf search and proposed limited lazy wolf search for ranking attributes and the experimental results illustrates the framework achieving excellent performance than the antecedent. For demonstrating the stability of proposed framework in all problems, numerous experiments are carried out using prominent benchmark datasets and a proposed crime propensity prediction dataset. The crime propensity prediction accuracy accomplished by fine tuning the proposed algorithm was 99.3617% directing crime free society.