| dc.description.abstract | This thesis describes an investigation into the depth-dependent effects in proton radiation, evaluated by the use of Thermoluminescence Dosimetry (TLD) with CaSO4:Tm and LTB:Cu as target materials. Data collected on these target materials from experiments at Aarhus University and the University of Oslo with variable depth positions and dosages were preprocessed and inspected for general distribution and noise. The Linear Energy Transfer (LET) dependent Thermoluminescence (TL) response was investigated, showing a difference for intermediate energy protons between samples irradiated in the Bragg Peak and samples positioned early in the LET curve. A glow curve deconvolution was carried out, extracting TL parameters from the samples. Descriptive statistical features were also extracted from the samples. Statistical features, features extracted through deconvolution and raw glow curves were used to create predictive models of LET. Models based on Random Forest (RF), Principal Component Analysis (PCA) and Partial Least Squares (PLS) and Neural Networks (NN) were explored. Binary classification models gave satisfactory results between high and low LET with a validation/test accuracy of 0.93/0.87, and between samples irradiated in the Bragg peak and non Bragg peak with an validation/test f1-score of 0.96/1.0, while a reliable high precision regression model was not found. | |
