A Robust Ensemble Machine Learning Framework for Depth-Based Salt Zone Classification

Abstract

Salt distribution analysis is important for understanding subsurface conditions and
improving environmental and geological studies. Traditional methods for identifying
salt concentration zones are often time-consuming and may not perform well when
handling large datasets. To improve classification accuracy, this study proposes an
improved depth-wise salt zone classification method using an ensemble machine
learning approach that combines CatBoost and LightGBM algorithms. A Salt
Quantification dataset containing 2560 samples was used for model development.
Salt zones were classified into four categories based on salt coverage percentage: Low
Salt, Moderate Salt, High Salt, and Very High Salt. Data preprocessing techniques
such as feature engineering, normalization, and stratified train–test splitting were
applied to improve model performance and reduce bias.The ensemble model was
evaluated using five-fold cross-validation and standard performance measures.
Experimental results showed excellent classification performance with high
prediction consistency across all evaluation metrics. The proposed method
demonstrates that ensemble learning can be an effective and reliable approach for
automated depth-wise salt zone classification. This framework may support future
applications in environmental monitoring, subsurface analysis, and intelligent salt
quantification systems.