An Interpretable Sme Grouth Model Combaning With LightGBM Algorithm And Neural Network

A scientific and reasonable evaluation of the growth of an enterprise and analysis of its growth level and potential will help investors avoid blind investment and improve the scientific nature of investment.At the same time,it can also help business managers optimize decision-making behaviors and improve economic benefits.The traditional method of evaluating the growth ability of a company generally uses statistical methods.This method uses fewer data dimensions and a simple evaluation model,which makes it difficult to solve the problem of complex enterprise growth evaluation.Therefore,based on the data of small and medium-sized enterprises,this thesis proposes an interpretable enterprise growth evaluation model that combines machine learning and text mining technology.The construction of the corporate growth model can be divided into three steps.First,use web crawlers to capture public opinion and public opinion information on the Internet,and combine public opinion and public opinion information in the Cathay Pacific database(economic and financial research database)as public opinion data.These public opinion data are analyzed through the pre-training model combined with fine-tuning,and the public opinion-related features are constructed.Then,obtain the financial data about the enterprise from the Guotaian database and process it to extract more effective information.Input the processed features of financial data and related features constructed by public opinion into LightGBM to obtain new features.This new feature will also be the input of the subsequent neural network.Finally,a special neural network is used to output a score that can represent the growth of the company,and at the same time a set of parameters that can explain the growth of the company are obtained.Compared with the traditional statistical method,the combination of the neural network and the Light GBM algorithm can more effectively extract the features in the data and mine more information.The final results also show that the model has better performance than traditional statistical methods.Compared with general machine learning methods,the model is more interpretable through a reasonable hierarchical structure,and its performance is also slightly higher than that of general models.In addition,by expanding the data dimension,supplementing and analyzing public opinion data,and adding it to the final modeling,the dimension of enterprise growth evaluation is increased,and the performance of the model is also improved.