改进利益驱动神经网络在欺诈信息的应用研究

为了研究个体收益和代价实现总体净收益的最大化问题，提出了利益驱动的人工神经网络（ANN）分类方法。该方法引入了惩罚函数，根据实例不同的重要程度对不同实例的误分类给予可变惩罚，并在之后对净利益进行最大化处理。为了生成对个体的惩罚，参照每个实例的收益，通过改变函数值对误差平方和函数进行了修改，提出了七个不同版本的ANN模型。两个欺诈信息的实验结果表明，与原ANN、决策树和朴素贝叶斯分类器相比，所提模型的不同版本在净利润项上的性能优于其他方法，而且能够针对不同的数据集采用不同的权值生成方式。

Application research of improved interest-driven neural network in fraud information

To study the problem of maximizing individual income and cost to achieve total net income, this paper proposed an interest-driven artificial neural network(ANN) classification method. In this method, it introduced penalty function, and gave variable penalties for misclassification of different instances according to the importance of different instances, and then maximized the net benefits. In order to generate penalties for individuals, according to the benefits of each instance, it proposed seven different versions of ANN models by modifying the sum of squares of errors function to change the value of the function. Compared with the original ANN, decision tree and naive Bayesian classifier, the experimental results of two fraudulent information show that different versions of the proposed model outperform other methods in terms of net profit items, and they can generate different weights for different data sets.