A Weighted Naive Bayes for Image Classification Based on Adaptive Genetic Algorithm

Naive Bayes (NB) is a simple and widely used classification model, but due to the conditional independence assumption, the accuracy of NB is not very competitive in the field of image recognition. Therefore, this paper proposes a Weighted Naive Bayes classification algorithm with an Adaptive Genetic Algorithm (AGA_WNB), which is used to reduce the impact of this assumption. First, reduce the dimensionality of the image and binarize the image. Then, the initial weights of the features are used as the initial population, and the classification accuracy of the Weighted Naive Bayes (WNB) model is used as the fitness function. Adjust the crossover probability and mutation probability according to the fitness function, and select the better chromosome to enter the next generation. Finally, the optimal weights are selected by iteration. The experimental results on the public dataset MNIST show that under the same environment, the average accuracy of AGA_WNB is 3.25% higher than Weighted Naive Bayes based on Genetic Algorithm (GA_WNB) and 9.7% higher than NB. The single digit accuracy of AGA_WNB is 19% higher than NB. Compared with the comparison methods, the accuracy of AGA_WNB is also improved, and it has a good application prospect.