Enhancing Minerals Prospects Mapping with Machine Learning: Addressing Imbalanced Geophysical Datasets and Data Visualization Approaches

• Published in: 2023 34th Conference of Open Innovations Association (FRUCT) Vol. 34; no. 1; pp. 125 - 135
• Main Authors: Nidhi, Dipak Kumar, Seppa, Iiro, Farahnakian, Fahimeh, Zelioli, Luca, Heikkonen, Jukka, Kanth, Rajeev
• Format: Conference Proceeding Journal Article
• Language: English
• Published: FRUCT Oy 15.11.2023; FRUCT
• Subjects: class imbalance problem; Data visualization; Logistic regression; Machine learning; mineral prospectivity mapping; multilayer perceptron; Multilayer perceptrons; parallel coordinate plot; principle component analysis; smote; Technological innovation; Training data
• ISSN: 2305-7254; 2305-7254; 2343-0737
• DOI: 10.23919/FRUCT60429.2023.10328164

Minerals prospects mapping plays a pivotal role in the sustainable development of mineral resources, offering critical insights into subsurface geology and mineral potential. Traditional geological methods are often labor-intensive and time-consuming. In contrast, machine learning (ML) techniques have emerged as a powerful tool for accelerating and improving the accuracy of mineral prospect mapping. This article explores an innovative approach to enhance the performance of supervised ML models, specifically logistic regression and multilayer perceptron. One of the primary challenges in mineral mapping is dealing with imbalanced geophysical datasets, where positive samples (indicating mineral occurrences) are vastly outnumbered by negative samples (non-mineral areas). This imbalance can lead to biased model predictions, favoring the majority class while neglecting the minority class. To address this issue, we propose a novel oversampling technique that generates synthetic samples for the minority class, effectively rebalancing the dataset. By introducing diversity to the training data, our approach mitigates the bias and enhances the models' ability to identify mineral prospects accurately. The proposed approach empowers ML models to discriminate between mineral-rich and non-mineral areas with unprecedented precision, facilitating more informed decision-making for resource exploration and extraction. Ultimately, the integration of imbalanced dataset handling and data visualization techniques offers a robust framework for harnessing the potential of machine learning in mineral prospect mapping.