Publication
Economic Forecast of the Southern China on BP Neural Network — Taking Chongqing as an Example
Authors: Wenke Du, Jing Ge, Shidong Sun
This study explores the economic development patterns of Chongqing and southern China using a Back Propagation Neural Network (BPNN) model. It constructs comprehensive regional economic indicators via a sampling methodology, and applies PCA and time-series dimensionality reduction techniques to train and validate the forecasting model.
Using data from 2000 to 2019 obtained from the National Bureau of Statistics, financial reports, and regional yearbooks, the model effectively predicts regional GDP trends and reveals meaningful correlations with national development. The proposed framework provides a replicable mechanism for policy-driven regional economic analysis.
Application of Support Vector Regression in Prediction Model Using Genetic Algorithm Optimization
Authors: Wenke Du, Ruihan Chen, Zhenglong Cong
Housing prices have long been a central issue in public economics, yet existing forecasting models often fail to adapt across regions and time periods due to the complex, multifactorial causes of price variation. This study focuses on enhancing the predictive power of Support Vector Regression (SVR) by optimizing key model parameters.
A genetic algorithm is used to tune penalty coefficients, kernel parameters, and insensitive loss functions within the SVR framework. The optimized SVR model is then applied to simulate and predict housing price trends.
Results from model simulation demonstrate that the genetic algorithm–optimized SVR model shows significant improvement in both convergence speed and predictive accuracy, validating its effectiveness and feasibility for real-world housing price forecasting applications.
Tokyo Stock Exchange Prediction with a Hybrid Model of LightGBM and DNN
Authors: Y. Yang, X. Zhang, S. Liu, W. Du
With the growing popularity of stock investment as a mainstream wealth management strategy, predicting stock prices has become a key research topic. This study focuses on the JPX Tokyo Stock Exchange prediction task using data provided by the Kaggle platform.
The proposed model combines the strengths of LightGBM and Deep Neural Networks (DNN) to build a hybrid predictive framework. The Sharpe Ratio is used as the primary performance metric to evaluate risk-adjusted return quality.
Experimental results show that the hybrid model achieves the best performance, yielding a Sharpe Ratio of 0.152—outperforming standalone XGBoost, LightGBM, and DNN by margins of 0.041, 0.032, and 0.004, respectively.
In general,these papers have also been published on Atlantis Press, Scopus, SCI(Web of Science), IOPScience, Journal of Humanities, Arts and Social Science, EI(Engineering Village Statistics Basement), CPCI(Conference Proceedings Citation Index) etc.and they’ve also been included in ICFIED, IOPSCIENCE, EUDL(European Union Digital Library),SEMANTIC SCHOLAR, CNKI and other multiple platforms include the magazine writing
You can find my articles on my Google Scholar profile but they are also listed below with links in some cases.
Working Papers
- RDD Data Mining about Status Quo of Russian social welfare——based on Salary Difference Related to Gender, Age, Degree etc
- Progress and Practices of Internationalization of Chinese Multinational Enterprises (MNEs)