An artificial intelligence powered financial relationship analysis based on planetary positions and aspects

This doctoral research presents a systematic, comprehensive, and data-driven framework for examining the potential relationships between planetary motions and financial market indicators through artificial intelligence and deep learning methodologies. The primary aim of the study is to assess the conditions under which astronomical parameters, when integrated with financial time series, may exhibit meaningful patterns and to evaluate whether such patterns can be represented by modern machine learning models. High-resolution astronomical data obtained from the NASA JPL Horizons system were programmatically processed; planetary and asteroid positions, angular relationships, retrograde motions, zodiac sign transitions, lunar cycles, and eclipse events were encoded in a time-series format. These astronomical features were then synchronized with financial market data from 19 instruments exhibiting diverse market dynamics, including BIST, Bitcoin, Crude Oil, DAX, DJI, U.S. Dollar, Ethereum, Euro, FCHI, FTSE, Gold, Japanese Yen, Litecoin, NASDAQ, NYSE, Russell, Silver, S&P, and SSE, thereby constructing a comprehensive interdisciplinary dataset. The dataset was optimized using advanced feature selection algorithms, including Mutual Information (MI), Recursive Feature Elimination (RFE), Principal Component Analysis (PCA), and the Granger Causality Test, enabling a structured evaluation of potential relationships between astronomical and financial variables from both an information-theoretic and causality-oriented perspective. Subsequently, extensive relationship analyses were conducted using four deep learning architectures (ANN+MERT, RNN+MERT, LSTM+MERT, GRU+MERT), achieving notable predictive performance across different market indicators. A key contribution of this dissertation is the implementation of two test scenarios constructed under real-time chronological conditions. In the first scenario, the direction of daily volatility (increase/decrease) was modeled as a classification problem; in the second scenario, daily volatility levels were examined using regression models. Both scenarios were implemented using strictly time-ordered data splits, ensuring realistic forward-looking evaluation. Across several markets, the selected astronomical feature set contributed to identifiable patterns associated with volatility behaviour under specific conditions.To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was applied, enabling global and local assessment of feature contributions and increasing transparency regarding the decision mechanisms of the models. This interpretive layer provided a systematic view of how, and under which circumstances, astronomical variables may provide informative signals to machine learning systems. Overall findings indicate that when processed and integrated appropriately, astronomical data can produce meaningful patterns alongside financial time series and may serve as an alternative source of information within advanced AI-driven modelling frameworks. The study offers an interdisciplinary and scalable research architecture, positioning astronomical data modelling within the broader domains of computational finance, artificial intelligence, and time-series analysis, thereby contributing an original perspective to the academic literature.