ABSTRACT The discovery of archaeological sites traditionally entails the utilisation of physically demanding exploration methodologies, including terrain surveying and the analysis of historical records. Recent technological developments have led to an increased use of non‐invasive remote sensing techniques, including Google Earth, LiDAR and aerial photography, in southern Africa. The application of machine learning (ML) and deep learning (DL) techniques is becoming increasingly prevalent in the field of archaeology. However, their utilisation remains constrained in southern Africa due to the inherent complexity of the subject matter. This study assesses the efficacy of automated machine learning (AutoML) tools for the identification of stone walling in Kweneng, a Late Iron Age urban settlement in the southern Gauteng region of South Africa. The study aims to identify the most appropriate ML models and algorithms, considering landscape variables such as aspect, elevation and slope. The findings demonstrate that the LightGBM algorithm is the most efficacious for detecting stone walling, with elevation being a pivotal factor for site detection. High‐resolution images enhance model performance by emphasising local context. The study emphasises the necessity to consider diverse ML modes to optimise algorithms, marking a preliminary step towards using ML and DL in archaeological site prediction. These insights can markedly enhance the accuracy and efficiency of archaeological research, offering a more profound understanding of past landscapes and human activities.
Mncedisi Siteleki (Wed,) studied this question.