Causal inference and forecasting in the mining industry: Applications of econometric, cointegration, wavelet coherence, bayesian, and machine learning methods

The application of causality analysis in the mining industry is crucial for enhancing decision-making, optimizing operational efficiency, and improving risk management, yet it remains an underexplored area in current research. This thesis, through four journal papers, addresses the pressing need for advanced causal inference techniques that move beyond traditional correlational models, which often fail to capture the complex, dynamic relationships inherent in the industry. By introducing a range of statistical, econometrical methodologies, and Bayesian methods, this research offers a comprehensive framework for understanding and applying causality, contributing to more informed and strategic decision-making in mining, mineral processing and metallurgy operations.The first study presents a novel application of cointegration and causality testing within the mining industry, offering a foundational framework for understanding long-term equilibrium relationships between critical factors such as commodity prices, production output, and market demand. By employing Granger causality, Variable Lag Granger Causality and Johansen cointegration tests, the study reveals the directionality and magnitude of causal relationships, allowing for a more accurate identification of the forces driving industry trends. The results highlight the limitations of relying solely on correlation-based analyses, demonstrating that causality-based approaches provide deeper insights into the underlying mechanisms governing the mining market.The second study focuses on the application of wavelet coherence and connectedness analysis to capture time-varying and frequency-dependent causal relationships. This method allows for a more granular understanding of how relationships between variables evolve over time, particularly in response to external shocks such as market fluctuations or geopolitical events. The case study illustrates the practical benefits of this approach, emphasizing its relevance in the context of the mining industry's inherent volatility.The third study presents Bayesian linear regression as a probabilistic framework for modeling causal dependencies in the mining sector, comparing it to machine learning methods. Unlike traditional linear models, the Bayesian approach incorporates prior knowledge and accounts for uncertainty in the estimation process. Case studies demonstrate a comparative analysis of Bayesian linear regression and random forest, both of which can be used for prediction. However, the Bayesian method also reveals the relationships among geological, plant, and mining variables, emphasizing its ability to capture more precise relationships and uncertainty as causes.The fourth and final study employs Bayesian hierarchical modeling to account for multi-level causal relationships within the mining industry. This method is particularly well-suited for sectors like mining, where different domains include distinct subdomains. By structuring the analysis to consider these various levels, the study offers a more comprehensive view of causality within the grinding process thereby facilitating decision-making and enhancing reliability processes. The Bayesian hierarchical model also incorporates uncertainty across multiple levels, making it an invaluable tool for long-term planning and causal analysis in mining operations.Through these studies, this thesis addresses the limitations of traditional analytical methods and introduces advanced causality techniques to tackle the complex challenges faced by the mining industry. The research results demonstrate the potential for causality-based approaches to bolster operational efficiency, enhance risk management, and yield more accurate market predictions. This thesis, therefore, significantly contributes to the growing body of knowledge on causality analysis in the mining sector and lays the groundwork for future research and innovation in this area

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.