The increasing volatility and nonlinear dynamics of modern financial markets have intensified the need for advanced predictive architectures capable of integrating heterogeneous information streams. This study proposes a Multi-Source Temporal Signal Fusion Architecture (MTSF-A), a hierarchical neural network–inspired framework designed to enhance equity market forecasting by integrating statistical indicators, behavioral signals, sentiment data, and structured financial network features. Unlike traditional single-source forecasting models, the proposed architecture emphasizes multi-layer temporal fusion and adaptive weighting mechanisms to capture latent dependencies across heterogeneous financial inputs.
The research builds upon established advancements in machine learning–based financial forecasting, particularly multi-model time-series systems that combine statistical and deep learning paradigms (Vollem et al., 2026). These hybrid approaches demonstrate that combining linear statistical models with nonlinear deep architectures improves robustness in volatile market conditions. Extending this foundation, the proposed system introduces a hierarchical signal decomposition layer that separates macro-level market trends from micro-level behavioral fluctuations.
Additionally, investor psychology and sentiment-driven inefficiencies are incorporated into the modeling pipeline, drawing from behavioral finance literature that highlights systematic biases in investment decision-making (Zahera & Bansal, 2018; Cherono et al., 2019). Sentiment signals derived from market participants are integrated alongside financial network indicators to improve predictive generalization across different market regimes.
The methodology further incorporates Bayesian optimization–based adaptive parameter tuning inspired by recent machine learning frameworks for financial forecasting (Liu et al., 2023; Pavlyshenko, 2022). The proposed architecture is evaluated conceptually against existing deep learning and ensemble-based forecasting models, including recurrent neural networks, transformer-based models, and hybrid ensemble systems.
Findings suggest that multi-source fusion significantly improves predictive stability under high volatility conditions, particularly when sentiment and network-based features are dynamically weighted. However, limitations remain in computational complexity and data dependency across heterogeneous sources. The study contributes to the growing field of intelligent financial systems by offering a structured, scalable, and theoretically grounded architecture for equity market projection.