Missing well logs prediction method based on K-nearest neighbors regression

In petroleum exploration, well logs are crucial for reservoir characterization. However, missing well logs frequently occur due to tool failures or economic constraints, which can impede accurate subsurface modeling. This research presents a method for predicting missing well logs using the K-Nearest Neighbors (KNN) regression algorithm, trained on data from the University of Kansas. The study focuses on predicting Delta T (DT) and Gamma Ray (GR) logs across five wells. The KNN model effectively captures the relationships between available and missing logs through proximity-based predictions, leveraging patterns from similar well-log data. The model achieved high test accuracy with R^2 scores ranged from 0.942 to 0.963 for DT and 0.927 to 0.930 for GR, indicating robust performance and generalization of unseen data. Observations of minor overfitting were noted, with training accuracy slightly exceeding test accuracy; however, these differences do not significantly detract from the model’s effectiveness. The results demonstrate that the KNN algorithm is a promising method for estimating missing well logs, effectively enhancing reservoir characterization workflows in data-limited scenarios.