evaluation of band ratio technique for prediction of iron-titanium mineralization using ensemble machine learning model: a case study from khamal area, western saudi arabia

Innovation in mineral exploration occurs either in the construction of new ore deposit models or the development of new techniques used to locate the ore deposits. band ratio is the image processing technique developed for mineral exploration. the present study presents a new approach used to evaluate the band ratio technique for discrimination and prediction of the iron-titanium mineralization exposed in the khamal area, western saudi arabia using the ensemble random forest model (rf) and spot-5 satellite data. spot-5 band ratio images are prepared and used as the explanatory variables. the target variable is prepared in which (70%) of the target locations are used for training and the rest are for validation. a confusion matrix and the precision-recall curves are constructed to evaluate the rf model performance and the receiver operating characteristics curves (roc) are used to rank the band ratio images. results revealed that the 3/1, 2/1 3/2 band ratio images show excellent discrimination with auc values of 0.986, 0.980 0.919 respectively. the present study successfully selects the 3/1 band ratio image as the best classifier and presents a new fe-ti mineralization image map. the present study proved the usefulness of the random forest classifier for the prediction of the fe-ti mineralization with an accuracy of 97%.