Chi-square Test Neural Network

Artificial neural network(ANN)has been a research hotspot in the field of artificial intelligence since the 1980s.It abstracts the neural network of human brain from the perspective of information processing,builds some simple model,and forms different networks according to different connection modes.It is composed of a large number of nodes(or neurons)connected to each other.Chi-square test is a widely used hypothesis testing method,which is mainly used in statistical inference of classified data.It determines the chi-square value by calculating the deviation between the actual observed value and the theoretical inferred value.This paper introduces the chi-square test neural network:a single hidden layer backpropagation neural network using chi-square test theorem to redefine the cost function and the error function.The weights and thresholds are modified using standard backpropagation algorithm.In the algorithm,a small number of labeled samples are divided into some subinterval,and the theoretical probability of a certain type of samples within each subinterval is calculated as a reference,and the training model is trained with unlabeled data.Due to the homogeneity of chi-square test,the prediction probability that the classification result of the input samples fall in each subinterval during training should be approximate to the theoretical probability.The two probabilities are substituted into the chi-square test statistical formula and compared with the critical value to judge whether the training is over.If not,update the parameters according to the improved error formula and continue to train the model.Therefore,the algorithm proposed in this paper can use a small number of labeled samples as a reference and use a large number of unlabeled samples to train the model.The proposed approach was tested on five sets of public and data sets provided by UCI machine learning library and private apple near-infrared spectral data and lung CT images.The proposed approach has the advantage of making consistent data distribution over training and testing sets.It can be used for binary classification.The experimental results on real world data sets indicate that the proposed algorithm can significantly improve the classification accuracy comparing to related approaches.This proves that the algorithm is feasible.Moreover,only a small amount of labeled data is needed to calculate the theoretical probability as a reference.And to a certain extent,it solves the problem of difficult data acquisition with labels.