Modified Extreme Learning Machine Algorithm with Deterministic Weight Modification for Investment Decisions based on Sentiment Analysis

Abstract

A significant problem in economics is stock market prediction. Due to the noise and volatility, however, timely prediction is typically regarded as one of the most difficult challenges. A sentiment-based stock price prediction that takes investors' emotional trends into account to overcome these difficulties is essential. This study aims to enhance the ELM's generalization performance and prediction accuracy. This article presents a new sentiment analysis based-stock prediction method using a modified extreme learning machine (ELM) with deterministic weight modification (DWM) called S-DELM. First, investor sentiment is used in stock prediction, which can considerably increase the model's predictive power. Hence, a convolutional neural network (CNN) is used to classify the user comments. Second, DWM is applied to optimize the weights and biases of ELM. The results of the experiments demonstrate that the S-DELM may not only increase prediction accuracy but also shorten prediction time, and investors' emotional tendencies are proven to help them achieve the expected results. The performance of S-DELM is compared with different variants of ELM and some conventional method.