Stock Market Prediction: Knowledge Gaps, Methods and Experimental Analysis

Academics and practitioners struggle to forecast financial market behavior. Recent years have seen remarkable growth in ML and DL research. Twenty-five open access articles from 2024 to 2026 are analyzed to identify knowledge gaps and suggest experiments to address them. Innovative models such LSTM networks with symbolic genetic programming, transformer architectures with generative decoding, CNNs for chart analysis, reinforcement learning for trading decisions, and privacy-preserving federated learning frameworks are reviewed. These studies have small, homogeneous datasets, weak external factor inclusion, low interpretability, and inadequate long-term or online forecasting despite their diversity. To fill these deficiencies, we study time–frequency analysis, sentiment integration, incremental learning, hybrid ensemble modeling, and federated learning. Experimental examples using synthetic financial time series data and three machine learning models (linear regression, random forest, and multi-layer perceptron) demonstrate how modeling choices affect prediction accuracy. The report concludes with challenges and prospects.