CAJ | 학술논문

现有的贝叶斯网络模型选择的各种评价准则中，评价准则最为常用。然而在大样本前提下近似推导得到的准则，在基于小样本条件下进行贝叶斯网络模型选择时，得到的网络过于稀疏，网络连通性不足。而准则在小样本条件下学习贝叶斯网络，选择出的网络往往比较复杂。针对这一情况，本文通过引入调整因子，将与准则的惩罚项进行加权，提出了评价准则，从而提高了小样本情况下贝叶斯网络模型选择的效果。理论分析和实验结果反映了改进的准则的合理性和有效性。
현유적패협사망락모형선택적각충평개준칙중，평개준칙최위상용。연이재대양본전제하근사추도득도적준칙，재기우소양본조건하진행패협사망락모형선택시，득도적망락과우희소，망락련통성불족。이준칙재소양본조건하학습패협사망락，선택출적망락왕왕비교복잡。침대저일정황，본문통과인입조정인자，장여준칙적징벌항진행가권，제출료평개준칙，종이제고료소양본정황하패협사망락모형선택적효과。이론분석화실험결과반영료개진적준칙적합이성화유효성。
TheBⅠC is the most commonly used one among the existing evaluation standards for Bayesian net-work model selection. However,under the Precondition of the large samPle,the BⅠC criterion of aPProximation is obtainedon the basis of the Bayesian networks,which leads to the sParse network and the in sufficient network connectivityunder the condition of the small samPle model,whilethe AⅠC criterion under the condition of the small samPle leads to the comPlicated selection. Ⅰn order to solve this situation,in this PaPer,by introducing the adjust-ment factor,the BⅠC Punishment with the weighted AⅠC criterion,ProPoses the evaluation criterion,so as to im-Prove the small samPle cases,the effect of the Bayesian network model selection. Theoretical analysis and exPeri-mental results reflect the rationality and validity of the imProved criterion.