Multi-source data fusion for economic data analysis

  • PDF / 1,104,901 Bytes
  • 11 Pages / 595.276 x 790.866 pts Page_size
  • 30 Downloads / 209 Views

DOWNLOAD

REPORT


(0123456789().,-volV)(0123456789(). ,- volV)

S.I.: HIGHER LEVEL ARTIFICIAL NEURAL NETWORK BASED INTELLIGENT SYSTEMS

Multi-source data fusion for economic data analysis Menggang Li1,2,3 • Fang Wang2,4 • Xiaojun Jia1,3 • Wenrui Li2,4 • Ting Li4 • Guangwei Rui4 Received: 13 August 2020 / Accepted: 11 November 2020  Springer-Verlag London Ltd., part of Springer Nature 2020

Abstract Economic data include data of various types and characteristics such as macro-data, meso-data, and micro-data. The source of economic data can be the data related to economy held by the National Bureau of statistics and a various software. These multi-source and heterogeneous data have important value for economic analysis and forecasting. Taking into account the limitations of existing methods such as low accuracy and complex calculations, this paper proposes an economic data analysis and prediction method based on machine learning. We use machine learning to solve the data fusion problem in the process of multi-source data analysis and prediction in the economic field. Specifically, we proposes an economic data analysis and forecasting method combining convolutional auto-encoder and extreme gradient boosting algorithms. This method uses a convolutional auto-encoder to extract the data characteristics of the normalized parameter sequence and uses it to train an extreme gradient boosting model to predict the level of economic development and evaluate the importance of each influencing factor. Finally, through a case study, this paper integrates the data of labor force, education and population to forecast GDP. Through the verification of this case, the prediction accuracy of the proposed method is higher than the AE-XGBoost method and CAE-1D-XGBoost method used in this experiment, and the error is kept below 11.7%. Keywords Economic forecasts  Multi-source data fusion  Convolutional auto-encoder  Feature extraction  Extreme gradient boosting

1 Introduction

& Fang Wang [email protected] & Wenrui Li [email protected] Menggang Li [email protected] 1

National Academy of Economic Security, Beijing Jiaotong University, Beijing, China

2

Beijing Laboratory of National Economic Security Earlywarning Engineering, Beijing Jiaotong University, Beijing, China

3

Beijing Center for Industrial Security and Development Research, Beijing Jiaotong University, Beijing, China

4

School of Economics and Management, Beijing Jiaotong University, Beijing, China

At present, economic forecasting methods can be divided into two categories: quantitative research and qualitative research [1]. Quantitative methods mainly use historically accumulated data, use a certain mathematical model to model and analyze these data, and give concrete values for the prediction results. Quantitative methods are divided into four categories: basic forecasting, time series, econometric model forecasting, and input–output analysis forecasting [2]. Among them, the basic forecasting method in the quantitative analysis method contains very simple calculations, such as