CAJ | 학술논문

集成学习是提高分类精度的一种有效途径,已在许多领域显示出其优于单个分类器的良好性能.观察学习是一种基于社会学习理论的集成学习方法,以往对其研究集中于同构模式.在此提出了基于异构模式的观察学习策略,通过训练、观察、再训练三个阶段完成学习.在UCI标准数据集上对异构环境下的观察学习算法进行了实验研究.结果表明,该方法优于多数投票法和单个分类算法,其对弱分类器组成的分类器集合尤其有效.从偏差/方差分解的角度对观察学习提高分类性能的原因进行了论证,结果表明,观察学习算法既可以降低偏差,也可以降低方差.
집성학습시제고분류정도적일충유효도경,이재허다영역현시출기우우단개분류기적량호성능.관찰학습시일충기우사회학습이론적집성학습방법,이왕대기연구집중우동구모식.재차제출료기우이구모식적관찰학습책략,통과훈련、관찰、재훈련삼개계단완성학습.재UCI표준수거집상대이구배경하적관찰학습산법진행료실험연구.결과표명,해방법우우다수투표법화단개분류산법,기대약분류기조성적분류기집합우기유효.종편차/방차분해적각도대관찰학습제고분류성능적원인진행료론증,결과표명,관찰학습산법기가이강저편차,야가이강저방차.
Ensemble method has shown the potential to increase classification accuracy beyond the level reached by an individual classifier alone.Observational Learning Algorithm (OLA) is an ensemble method based on social learning theory.Previous work mainly focused on OLA for homogeneous ensembles,such as neural networks ensembles.In this paper,OLA for heterogeneous ensembles was proposed,which is a process with three steps:training,observing,and retraining.Experiments on five datasets from the UCI repository show that,OLA outperforms the individual base learner and majority voting when base learners are not capable enough for the given task.Bias-variance decomposition of the error indicates that OLA can reduce both bias and variance.