A Hybrid Forecasting System Based on Comprehensive Feature Selection and Intelligent Optimization for Stock Price Index Forecasting

Accurate forecasts of stock indexes can not only provide reference information for investors to formulate relevant strategies but also provide effective channels for the government to regulate the market. However, due to its volatility and complexity, predicting the stock price index has always been...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Mathematics (Basel) Jg. 12; H. 23; S. 3778
Hauptverfasser: He, Xuecheng, Wang, Jujie
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Basel MDPI AG 01.12.2024
Schlagworte:
Accuracy
Algorithms
Analysis
Data compression
Efficiency
Forecasting
Forecasts and trends
Gaussian process
interval prediction
Investments
Machine learning
Multivariate analysis
Optimization
Political aspects
Prices and rates
recursive feature elimination with cross-validation
Root-mean-square errors
stack autoencoder
Statistical analysis
Stock exchanges
stock index forecasting
Stock price indexes
Stock prices
Stocks
Support vector machines
Trends
Variables
Volatility
ISSN:2227-7390, 2227-7390
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Accurate forecasts of stock indexes can not only provide reference information for investors to formulate relevant strategies but also provide effective channels for the government to regulate the market. However, due to its volatility and complexity, predicting the stock price index has always been a challenging task. This paper proposes a hybrid forecasting system based on comprehensive feature selection and intelligent optimization for stock price index forecasting. First, a recursive feature elimination with a cross-validation (RFECV) algorithm is designed to filter variables that have a significant impact on the target data from multiple datasets. Then, the stack autoencoder (SAE) algorithm is constructed to compress the feature variables. At last, an enhanced least squares support vector machine (LSSVM) algorithm is established to obtain high-precision point prediction results, and the Gaussian process regression (GPR) algorithm is used to obtain reasonable interval prediction results. Taking the Shanghai Stock Exchange (SSE) as an example, the root mean square error (RMSE) and mean absolute percentage error (MAPE) of the model were 6.989 and 0.158%, respectively. In addition, the prediction interval coverage probability (PICP) is 99.792%. Through experimental comparison, the model shows high prediction accuracy and generalization ability.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2227-7390
2227-7390
DOI:10.3390/math12233778