Accurate analysis of well log data is critical for characterizing subsurface formations and making informed decisions in hydrocarbon exploration and production. However, real-world datasets frequently suffer from missing values due to tool malfunctions, adverse borehole conditions, or inconsistencies in historical records. This problem is particularly acute in industrial datasets such as those from Sonatrach, Algeria’s national oil company, where well logs are often incomplete, unstandardized, and highly heterogeneous across multiple wells. This study addresses this critical limitation in the context of Sonatrach’s strategic goal to modernize subsurface analysis using machine learning. Building on recent regional efforts that apply AI for property prediction and lithofacies classification across Algerian fields, this work identifies a key gap: the lack of robust methodologies for handling incomplete well log data. To fill this gap, we evaluate six imputation techniques, ranging from simple statistical methods (mean, median) to advanced machine learning algorithms (KNN, MICE, XGBoost, Random Forest), including a novel application of a Wasserstein Generative Adversarial Imputation Network (WGAIN). The models are assessed using a real-world, partially incomplete dataset provided by Sonatrach. To further boost imputation quality, sequential depth-wise sliding-window statistics are incorporated, leveraging the temporal structure of well logs. Results demonstrate that ensemble-based and WGAIN yield significant improvements in data completeness and downstream petrophysical prediction performance. This study represents the first application of WGAIN to well log data and provides a reproducible framework for enhancing log quality in resource-constrained settings.