重庆科技大学学报（自然科学版）2025，Vol.27Issue(3)：70-77,8.DOI:10.19406/j.issn.2097-4531.2025.03.007
基于KAN-BiLSTM模型的股票指数预测研究
Research on Stock Index Prediction Method Based on KAN-BiLSTM Model
摘要
Abstract
To address the limitations of current neural networks in terms of accuracy and generalization capability for long-term stock index prediction,a novel model(KAN-BiLSTM)that integrates a KAN(Kolmogorov-Arnold net-work)with learnable activation functions and a Bidirectional Long Short-Term Memory(BiLSTM)network is pro-posed.This model utilizes BiLSTM to extract bidirectional temporal features of stock data and enhances the model's expressive power through the strong nonlinear function approximation ability of KAN,thereby improving the overall prediction performance.Experiments are conducted on multiple long-term stock datasets,and the results show that KAN-BiLSTM model improves the prediction accuracy by an average of approximately 5.68％compared to existing BiLSTM model,and also demonstrates better generalization ability,verifying its effectiveness in stock index predic-tion tasks.关键词
神经网络/KAN模型/BiLSTM模型/长跨度股票数据Key words
neural network/KAN model/BiLSTM model/long-span stock data分类
信息技术与安全科学引用本文复制引用
赵涛,赵迎庆..基于KAN-BiLSTM模型的股票指数预测研究[J].重庆科技大学学报（自然科学版）,2025,27(3):70-77,8.基金项目
安徽省高校自然科学重点项目"跨社交网络用户识别技术研究"(KJ2021A0483) （KJ2021A0483）
