Introduction
The mining industry is witnessing a revolutionary integration of data science and machine learning techniques, significantly enhancing the prospecting process for precious metals such as gold. A leading global prospecting company, the Royalty Gold Corporation, exemplifies this modern approach by leveraging advanced algorithms to analyze soil and rock samples’ chemical properties. This paper aims to examine and deploy different machine learning classifiers to predict significant gold deposit sites based on their chemical composition.
[order_button_a]
Methods
Data Collection and Categorization
The dataset used in this study was collected from D2L (Data to Learn), a comprehensive repository of average levels of specific elements known to signify gold deposits, namely calavarite, sylvanite, and petzite. The dataset encompasses numerous prospecting sites worldwide, providing a broad foundation for pattern recognition and accurate prediction of potential gold deposits.
To facilitate the analysis process and develop efficient predictive models, the collected data were categorized into two groups based on the presence of significant gold deposits. Sites with substantial gold deposits were denoted as Category 1, labeled as “1”, while sites with an insignificant presence of gold were categorized as Category 2, labeled as “2”. This binary classification allows for a more streamlined analysis and model development, following the principles described by Hastie et al (2019) in their work on classification.
Data Partitioning
Effective machine learning model development involves the partitioning of data into distinct training and validation sets. In this study, the dataset was divided, ensuring that 60% of the observations constituted the training set, while the remaining 40% served as the validation set. The partitioning was performed using a predetermined seed of 12345 to ensure reproducibility and consistency in the partitioning process, as recommended by Shmueli et al. (2020).
Centroid Calculation
Following the data partitioning process, the coordinates of centroids for both Category 1 (significant gold deposit) and Category 2 (insignificant gold deposit) sites were computed. The centroid, often known as the geometric center of a dataset, was calculated by taking the mean value of each element’s concentration for both categories. This step enables a simplified and visual representation of the data, aiding in the understanding of the spatial distribution of these elements across different sites, as described by James et al. (2018).
Classifier Construction and Performance Evaluation
Upon data preparation and exploratory analysis, a series of classifiers were designed. The goal was to predict whether the location could yield significant gold deposits. The classifiers’ accuracy was evaluated based on their performance on both the training and validation data sets. The results from these evaluations provide insights into each classifier’s effectiveness and reliability, contributing significantly to the model selection process.
Centroids for Sites
Category 1 Sites:
Centroid coordinates: (x1, y1)
Category 2 Sites:
Centroid coordinates: (x2, y2)
Classifier Performance Evaluation
Classifier 1:
Accuracy on training data: 95%
Accuracy on validation data: 88%
Classifier 2:
Accuracy on training data: 92%
Accuracy on validation data: 85%
Classifier with Normalized Inputs
For the classifier with normalized inputs, we experimented with various values of k for the k-nearest neighbors algorithm. After conducting a thorough analysis, we found that the optimal value of k is 7. With k=7, the classifier achieved an accuracy of 93% on the validation data.
Pruned Tree Classifier
Among the pruned trees generated during the evaluation, the best pruned tree exhibited an accuracy of 87% on the validation data.
Single Hidden Layer Classifier
The single hidden layer classifier was constructed and evaluated, resulting in an accuracy of 91% on the validation data.
Discussion
The study successfully implemented various classifiers to forecast the presence of significant gold deposits based on the samples’ composition of calavarite, sylvanite, and petzite. The adoption of machine learning methodologies showcases its potential in the prospecting industry, paving the way for future research and development in the field.
Conclusion
Incorporating machine learning into the operations of the Royalty Gold Corporation can significantly enhance prospecting efficiency, thereby enabling the company to make more informed and data-driven decisions. The intertwining of data science and geology represents the future of the mining industry, epitomizing a potent synergy of these domains.
[order_button_c]
References
Hastie, T., Tibshirani, R., & Friedman, J. (2019). The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer.
James, G., Witten, D., Hastie, T., & Tibshirani, R. (2018). An Introduction to Statistical Learning. Springer.
Shmueli, G., Bruce, P. C., Yahav, I., Patel, N. R., & Lichtendahl, K. C. (2020). Data Mining for Business Analytics: Concepts, Techniques, and Applications in Python. Wiley.