With the opening of the Growth Enterprise Market and the normalization of stock exchange approval by companies on the stock exchange,the number of Chinese A-share listed companies has rapidly increased.The registration system to be implemented will create extremely convenient conditions for companies to go public.Therefore,while providing investors with many excellent companies to obtain income,it also brings some difficulties for non-professional investors to choose a suitable listed company for investment.Many scholars conduct performance evaluation and classified research on listed companies based on the released financial reports and the information disclosed.At present,data mining based on Internet technology and K-Means algorithm is an effective method to solve this problem.It can mine hidden and difficult-to-discover relationships between financial data and overcome the dependence and limitations of traditional financial statistical report analysis,but K-Means clustering algorithm has a strong dependence on the selection of initialization conditions and is easy to fall into local optimum.Therefore,this thesis studies these problems and improves the K-Means algorithm to mine the internal links and differences between listed companies’ financial indicators and classify listed companies.It mainly includes the following two innovations:1.Propose an improved K-Means clustering algorithm for performance evaluation of listed companies.The algorithm first selects the initial clustering center based on the stepwise dichotomy of K-Means,and in the initial process of selecting the initial centroid,selects the sample point of the known initial maximum value as the centroid of the dichotomy,and the subsequent The experimental results brought into the sample set show that this kind of processing method is better than the clustering effect of this kind of algorithm.Using this method to analyze the financial statements of listed companies and evaluate the operating conditions can better help the decision-making and investment of fund companies.The investors choose which listed companies to invest in for reference.2.Propose a PSO-RF method for the classification of listed companies.After conducting cluster analysis on the financial data of listed companies,we then proceed with a classification study to classify the subsequent other listed companies accordingto the financial indicator data of different years as accurately as possible.In this paper,the random forest algorithm and the BP neural network algorithm are used to verify the classification evaluation indexes of the above data,and the particle swarm optimization algorithm is used to optimize the parameters.The experimental results show that the random forest algorithm after particle swarm optimization can significantly improve the effect of company classification.This article uses data mining to evaluate and classify the performance of different listed companies.According to the internal relationship between financial indicator data,it is more scientific than manual classification.I hope that it can be more objective for relevant analysts,investors and decision makers.,Rational and in-depth understanding of the company’s operating status to make timely and appropriate decisions.