Predicting the Archaeological Landscape : Archeological Density Estimation around the Ostlänken railroad corridor
Anna Karin Nilsson · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2016
This project aims to create a prediction model that predicts where it is likely to find archaeological features in an area, using machine learning methods. It uses the data from previous archaeological excavations together with geological data of the area as training data for the model. The model built consists of Kernel Density Estimation, combined with Random Forest. The report examines the advantages of these, as well as other models. On the training data the model performs rather well, with a sensitivity of 76% and a precision of 37%. But when predicting all points over the whole map it makes a very unlikely prediction: 98% of the map is predicted to have findings. When the model is run on training data that is transformed to have a more even spread between positive and negative cases a better result is found. In conclusion this project is a step towards a good prediction model, but more work is needed.