Investigating the Impact of Feature Engineering Techniques on the Predictions of Well Log Attributes

Abstract Accurately predicting well log attributes such as porosity is essential for reservoir evaluation and optimizing resource recovery in the oil and gas sector. Traditional methods often struggle to capture the complex, non-linear relationships inherent in geological data. This study investigates the impact of ten distinct feature engineering techniques on the predictive performance of machine learning models developed for porosity estimation using well log data from the Volve oil field. Feature extraction and feature selection techniques were applied to transform and refine high-dimensional well log data. Dimensionality reduction methods, including Principal Component Analysis (PCA), and various selection methods, such as embedded and filter-based approaches, were explored to generate optimized feature sets. These sets were then used to train and evaluate multiple machine learning algorithms. The objective of this study is to determine which feature engineering techniques, based on their underlying assumptions, most effectively capture patterns in the input data and enhance the accuracy of porosity predictions. This work contributes to the broader understanding of how data preprocessing choices influence model robustness and performance, ultimately supporting improved reservoir characterization and efficient resource development.