This project will explore how machine learning can create simple, human-readable rules from complex financial data. Stock markets are governed by patterns, constraints, and external factors, but these rules are often hidden within noisy data and difficult to extract with traditional methods. The project will use a neural network with a symbolic layer, built from basic mathematical operators (e.g., sums, products, powers, logs). This enables the model to learn interpretable formulas that describe stock price dynamics, rather than relying on black-box predictions. Training will be guided to ensure that the resulting rules are accurate, stable, and understandable to financial analysts and decision-makers. A key focus will be applying this approach to stock price forecasting, combining historical time series data with market indicators. The project builds on recent research in explainable AI for decision-making and extends it to generate interpretable forecasting formulas that can support risk assessment, trading strategies, and regulatory compliance in financial applications.