Volatility-aware Multimodal Market Forecasting via OPTICS Clustering, Spider Web Hyperparameter Optimization, and Incremental LSTM Prediction

This paper presents a combined multimodal predictive model based on stock market forecasting exploiting volatility-aware adaptive sampling, the Optics, Spider Web Optimization (SWO), Ball-Tree construction along with Incremental LSTM prediction.Proposed framework effectively integrates all opportunities available in the market as quantitative information, including prices and volumes, technical analysis variables, and sentiment analysis and pivots them into one model that is mutually supported.To evaluate the proposed multimodal stock market forecasting model, this paper uses a consolidated Yahoo Finance dataset of data between January 1, 2010, and December 31, 2020.This data contains the price and the volume of the stocks and the technical indicators like RSI and MACD, the sentiment data obtained by analysing the financial news and social media.Ball-Tree Models increase the scalability of the framework providing nearest-neighbor searches with high-dimensional financial data, whereas the domain of market regimes is determined by the use of the OPTICS clustering method based on the densities.Given the pivotal hyperparameters, the SWO algorithm has maximum optimization to enhance the predictive performance.It is highly dynamic because the framework is dynamically updated in real-time to deal even with volatile periods.The results were found to be highly improved and precision, recall and F1-score have all been found 80% and this means that the superiority of the work with regard to the prediction accuracy is achieved over the current models and therefore, the framework is a powerful tool of making decisions regarding finance in real time.