ENHANCED STOCK PRICE PREDICTION USING CRNN, SENTIMENT ANALYSIS, IABC OPTIMIZATION, AND ADVANCED TEXTUAL AND TECHNICAL FEATURE EXTRACTION

Abstract

Stock price prediction is a very crucial yet discouraging task in financial analytics because of the complexity and nonlinearity of the market. The paper proposes a hybrid prediction model that combines Convolutional Recurrent Neural Networks (CRNN) with the Improved Artificial Bee Colony (IABC) optimization for improved accuracy. This model combines textual data from major stock forums and technical market indicators from the Shanghai Stock Exchange for effective stock price prediction. Preprocessing of the textual data for quality assurance: text cleaning, tokenization, stop word removal, stemming, and lemmatization. Feature extracting techniques include Bag-of-Words (BoW), N-grams, and Improved Term Frequency-Inverse Document Frequency (ITF-IDF) to represent in numerical form. As mentioned, a CRNN has performed sentiment analysis that takes convolutional layers for spatial feature extractions and recurrent layers in temporal dependency modeling. These classified sentiments are combined with technical indicators to form a unified feature set for stock price prediction. The IABC algorithm further optimizes the model by hyperparameter tuning and improving convergence. Experimental verification time intervals prove the superiority of the model over traditional methods. This hybrid approach has significantly improved the accuracy, scalability, and practical applicability for financial decision-making.