一种基于“独立性差”的分类学习新算法

从"独立性差"角度出发,提出了ISE准则下的"独立性差"估计新方法(difference of independence estimation,DOIE).从数学模型上证明该算法与单类SVM等价且可用于解决分类问题.当数据集规模较大时,该算法的优势在于可用较少样本点表示两数据集中样本点间的关系,在保证精度的前提下,提高运算速度.该算法还可应用于两数据集独立性判断、检测流数据分布改变点的位置.若退化为单类数据集,可应用于概率密度估计.Benchmark和UCI数据集上的实验表明,该算法具有较好的性能.

A new classification algorithm based on the difference of independence

From the perspective of difference of independence, a new method of independence estimation approximating was proposed based on ISE criterion, called difference of independence estimation algorithm (DOIE). From the mathematical model, it is verified that this algorithm is equivalent to one class SVM and can be used for classification. When the scales of data sets are large, the algorithm can use fewer sample points to represent the relationship between all sample points which come from the two data sets. In the premise of ensuring the accuracy, the algorithm can improve processing speed. This method can also be used to estimate mutual information, used to determine the data stream's change location when data distribution is changed. If single-class sample set involved in computing, the algorithm can be applied to estimate the probability density as RSDE algorithm. Finally, the algorithm's validity is experimentally verified by the benchmark data sets and UCI data sets.