Classification method for imbalanced problems based on cost local generalization error

Abstract

The invention discloses a classification method for unbalanced problems based on cost local generalization errors, which comprises the following steps: performing normalization processing on each dimension feature of data; and calculating misclassification costs of different types of samples according to statistical laws of data sets ; Construct a cost-sensitive multilayer perceptron neural network model; for each sample, calculate the stochastic sensitivity (ST‑SM) corresponding to its local generalization error; complete model training by combining cost sensitivity and stochastic sensitivity of local generalization error. The present invention effectively combines the advantages of local generalization error in improving generalization ability and the faster training and testing time of the method based on the algorithm level while avoiding the disadvantage of the data level method having too much influence on the data set distribution The advantages of the method improve the stability of the classification of the unbalanced dataset and the accuracy of the classification results of each category, and obtain a more reasonable classification decision boundary.

Description

technical field [0001] The invention relates to the field of unbalanced classification of data sets, in particular to a classification method for unbalanced problems based on cost local generalization errors. Background technique [0002] With the development of computer science and technology, machine learning has played an extremely important role in many fields. In recent years, the situation of unbalanced datasets cannot be ignored in most fields and has become a hindrance to the development of machine learning. For example, in a physical examination, the number of healthy people is generally greater than the actual number of patients. If a healthy person is misclassified as a sick person, it only needs to be tested again manually; however, if the disease of the patient is ignored, the patient will not receive timely treatment. Ordinary machine learning assumes that the data set distribution is balanced, and the loss of misclassification is the same. Therefore, the mo...

AI Technical Summary

Problems solved by technology

However, the backpropagation algorithm has a disadvantage. The sequence of data input has a great impact on the results of model training. Compared with the data input later, the data input first has less impact on the model.

Images