Data classification method

Abstract

The present invention discloses a data classification method. The method comprises the steps of: S1, obtaining training set samples for training a classifier, and averaging the obtained training set samples according to the number of iterations required for training to obtain a plurality of training subset samples; S2, based on the Adaboost algorithm, training each of the training subset samples respectively by using a plurality of weak classifiers, when each weak classifier performs training, selecting some training subset samples and some error samples obtained by the previous weak classifier to constitute and form a final training sample, and obtaining a final ADB strong classifier from various weak classifiers after completing training; and S3, using the trained ADB strong classifier to classify to-be-classified data, and outputting a classification result. According to the method disclosed by the present invention, the data during classification training is complete, the trainingdata can be prevented from multiplying and over-fitting, and the method has the advantages of a simple implementation principle, high classification efficiency and precision, and the like.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a data classification method. Background technique [0002] Data classification is to map data to specified categories. Adaboost (Adaptive Boostin, adaptive enhancement) is an adaptive data classification algorithm that trains different classifiers (weak classifiers) for the same training set, and then classifies these weak classifiers The classifiers are assembled to form a stronger final classifier (strong classifier). Its adaptation is: the wrong samples of the last weak classifier will be strengthened, and all samples after weighting will be used to train the next basic classifier again. At the same time, a new weak classifier is added in each round until a predetermined small enough error rate is reached, or a pre-specified maximum number of iterations is reached. The Adaboost algorithm has a strong recurrent learning ability and can better combine and strengthen we...

AI Technical Summary

Problems solved by technology

[0005] (1) If full samples are used for iterative training, after each iteration, the number of samples will increase exponentially, increasing the difficulty of training;

[0006] (2) If random sampling is used to form the corresponding weight ratio, some samples will be missed, resulting in incomplete training;

[0007] (3) For samples with repeated errors, the original algorithm will consistently increase the weight of the samples. If the sample is an outlier, it will cause the subsequent classifier to over-train the outlier, thus deviating from the actual data sample.

[0008] At present, the improvement of the classification algorithm mainly includes two methods: the way of improving the algorithm itself, and the way of combining and superimposing multiple algorithms. Among them, the way of improving the classification algorithm itself is usually through some characteristics of the algorithm itself. Improvements, such as adding discriminant formulas, integrating other algorithms, and improving the algorithm structure, but due to the general complexity of the machine learning algorithm itself, the improved algorithm is basically based on a specific application scenario and is not universal. There are problems such as difficulty in improving and redundant and complex algorithms; as for the combination method, since it will not disrupt the structural characteristics of its own algorithm, it can complement each other according to the characteristics of different algorithms, which has great advantages and is applicable However, the combination improvement method of Adaboost algorithm is usually a simple combination method at present, which does not consider the above-mentioned problems in the training process of Adaboost algorithm itself, and there will still exist such as the number of samples above will increase exponentially, training Difficulty or incomplete training and problems such as over-training

Images