研究了在正例和未标记样本场景下不确定样本的分类问题,提出了一种新的算法NNPU(nearest neighbor algorithm for positive and unlabeled learning)。NNPU具有两种实现方式:NNPUa和NNPUu。在UCI标准数据集上的实验结果表明,充分考虑数据不确定信息的NNPUu算法要比仅仅考虑样本中不确定信息均值的NNPUa算法具有更好的分类能力;同时,NNPU算法在对精确数据进行分类时,比NN-d、OCC以及aPUNB算法性能更优。 We study the classification of uncertain samples in positive and unlabeled samples and propose a new algorithm NNPU (positive neighbor unlabeled learning). NNPU has two implementations: NNPUa and NNPUu. The experimental results on the UCI standard dataset show that the NNPUu algorithm, which takes the uncertain information into full consideration, has better classification ability than the NNPUa algorithm, which only considers the average value of the uncertain information in the sample. Meanwhile, the NNPU algorithm classifies the accurate data , Better performance than NN-d, OCC and aPUNB algorithm.