Research And Improvenment Of RBF Multi-label Learning Algorithm

With the development of the times,artificial intelligence has entered people's lives,and multi-label learning has become a research hotspot in the field of artificial intelligence.According to the number of labels corresponding to the marked sample set,it can be divided into single mark and multi mark learning,and many classification problems in real life can be regarded as multi mark learning problems.The main task of multi-label learning is to build a multi-labeled classifier model based on a given sample dataset and label dataset,and to predict unknown instances in the sample set.However,the traditional multi-label learning has a major problem: only the multiple tags in the instance are identified,and the correlation between the tags is not fully utilized,thus affecting the evaluation of each index.The RBF multi-label neural network learning algorithm is one of the typical multi-label learning methods.This paper studies and improves the RBF multi-label learning algorithm.The main research contents are as follows:(1)For the traditional RBF multi-label learning algorithm to ignore the association between different tags,an output-optimized ML-RBF algorithm is proposed.In the clustering stage,the distance between different sample centers and all sample centers is used for correlation analysis,and clustering is performed twice.First,K-means clustering is performed on the same marker samples,and then K-means clustering is performed on all samples.At the same time,the segmentation basis function is designed,and the stepwise evaluation is performed in the above two clustering results,thereby obtaining the correlation value between the tags.(2)Proposed PSO-ML-RBF multi-labeling algorithm based on particle swarm optimization.This method does not start from the multi-label learning algorithm model,but improves the indicators of the multi-label learning algorithm by optimizing the weight of the neural network.Using the particle swarm optimization algorithm and the fuzzy c-means method,the central function parameters of the basis function are optimized in the ML-RBF algorithm,and the weights between the hidden layer and the output layer in the neural network in the ML-RBF algorithm are adjusted to achieve multi-label learning.Algorithm optimization.(3)Analyze and verify the output-optimized ML-RBF and PSO-ML-RBF through experiments.On multiple common datasets,the improved algorithm is in Hamming loss,one error rate,coverage rate,sorting loss and average.Good results have been achieved in terms of accuracy.