Using logistic regression to validate object detections made by AI from ALS survey data ; a case study on Finnish tar kilns

Abstract

Aerial laser scanning (ALS) has opened new avenues for Finnish archaeology, with its ability to bypass canopy cover which enables researchers to study large areas of the terrain and topography of Finland at once. The LiDARK pilot project made use of this and explored the uses of artificial intelligence (AI) in the semi-automatic identification of archaeological objects from ALS survey images, using tar kilns (Finn. tervahauta) as their main object of interest. This thesis explores a new validation method for these object identifications using an archaeological predictive model. The model used applies a logistic regression algorithm to evaluate the statistical probability of tar kiln occurrence, based on independent variables derived from theory-based assumptions about tar kilns and the environmental qualities affecting their spatial distribution. The resulting three models all show variable success, with one model returning an overall accuracy of 87%.