Well-log Dataset for Lithology Classification Using Deep Learning Models

Anonymized well-log dataset for lithology classification from 9 boreholes in a sandstone-type uranium deposit. The dataset contains 89,160 samples with 6 well-log curves (CAL, DEN, DT, GR, RT, SP) and facies labels (0: Medium Sandstone, 1: Mudstone, 2: Glutenite, 3: Siltstone, 4: Coarse Sandstone). Well identifiers have been anonymized to protect sensitive geological location information. The data is explicitly partitioned into a Training dataset (8 boreholes) and a separate Blind test well for objective model evaluation. Original well names, coordinates, and exact spatial descriptions have been completely removed. This dataset was compiled to support and evaluate hybrid deep learning architectures (specifically incorporating TCN, BiLSTM, and Transformer models) for geophysical well-log interpretation. It accompanies a scientific manuscript currently under peer review.