New mFOIL algorithm combining Boosting technology
Jun Luo · Jisuanji yingyong yanjiu · 2009
In machine learning,the Boosting technology will give each training example a different weight so that a learning algorithm can focus on those hard training examples.On the other hand,the algorithm mFOIL can perform well in learning first-order rules,but the best m-value is difficult to determine when used to estimate the accuracy of candidate clauses.In order to escape the dilemma,this paper presented a new learning algorithm called BoostmFOIL here,which combined the Boosting technology so that increased the accuracy of learned rules dramatically for any m-value.Besides,the weights of noise datum which were assumed to have been labeled were tweaked to be very small,compared with those of normal datum.This can reduce the risk from noise and thus enhance the robustness of the new algorithm.Experiments on some benchmark domains show that not only do the Boosted algorithm(BoostmFOIL) perform better than the unboosted version(mFOIL) for any random m-value,but also the former can improve the accuracy of learned rules in very noisy environment.So it is concluded that combining the Boosting technology in mFOIL can enhance its accuracy and robustness.