Disease factor extraction method based on improved PSO-BP neural network and Bayesian method

Abstract

The invention discloses a disease factor extraction method based on an improved PSO-BP neural network and a Bayesian method. The method comprises the following steps that a self-adaptive weight strategy is introduced to optimize particle swarm optimization PSO; a BP neural network is optimized and trained according to the improved particle swarm optimization PSO and disease factor data used for training; the disease factor data used for testing is input into the trained PSO-BP neural network for prediction to obtain output results of disease risk factors, namely a weight matrix between all neurons of the neural network; according to the output results, the weight of a relationship between input and output is obtained through a conversion formula and recorded as a prior probability; according to the prior probability, the disease risk factors are obtained by combining with a Bayesian formula. Compared with a disease extraction method in the current disease prediction field, the method is more accurate, more efficient, more reliable and more stable.

Description

technical field [0001] The invention is designed in the field of big data technology and medicine, especially a disease factor extraction method based on the improved PSO-BP neural network and Bayesian method. Background technique [0002] At present, for the risk factors of cardiovascular and other diseases, various medical journals have summarized the risk factors that may cause the disease based on clinical experience and theoretical knowledge, but these methods often have some drawbacks. On the one hand, most analysis and research often have sample The quantity and analysis factors are small, so that the final result does not have generalization ability. Human analysis requires a lot of experience and knowledge accumulation, which lacks authority and scientificity; on the other hand, the analysis method is single, mostly based on statistical methods and clinical research and medical science. Based on experience, when the number of samples is large enough and the disease ...

AI Technical Summary

Problems solved by technology

[0004] In the current artificial neural network, the traditional BP neural network has two shortcomings: 1) BP neural network is easy to converge to the local optimum and stop training, causing large errors in prediction; It is easy to fall into the local optimum, prone to the shortcomings of low recognition rate and low accuracy; 3) The design of BP neural network is generally determined based on expert knowledge and experience. If the selection is too large, it may lead to low training efficiency and poor network performance. , poor fault tolerance; if the structure selection is too small, the network may not converge; 4) The construction of the BP neural network model has no scientific theoretical basis for the selection of the number of hidden layers and the number of nodes. Some empirical formulas and continuous trials are finally determined, which may lead to greater network redundancy and a heavier learning burden

Images