Research on Predicting Enterprise New Quality Productivity Based on K-Means Clustering and BP Neural Network

Abstract

Firstly, using text mining methods and analyzing the frequency of words related to new quality productivity in enterprise annual reports, we calculated the new quality productivity index for each enterprise from 2007 to 2022. We constructed 25 predictive features from three aspects: enterprise digital transformation, ESG, and productivity. To improve the efficiency of machine learning methods, the Boruta algorithm was used for feature selection, resulting in 21 predictive features. To enhance the accuracy of enterprise clustering, principal component analysis (PCA) and UMAP methods were used for dimensionality reduction, with UMAP showing better results than PCA. Using the reduced-dimensional features, the K-means clustering algorithm was applied to cluster each enterprise. The BP neural network method was then used to predict the new quality productivity index, achieving an accuracy of 97.7%. The study concluded: 1) "Technology" appeared most frequently (544,610 times) in annual reports, underpinning new quality productivity development; 2) Digital transformation, ESG, and productivity significantly impact new quality productivity; 3) Enterprises with longer establishment times have higher productivity indexes, promoting high-quality development; 4) Financial characteristics and ESG indicators can predict a company's level of new quality development.