TY - THES ID - 134741523 TI - Depression and Burnout Diagnosis by combining Data-Based Mechanistic Modelling and Machine Learning Methods AU - Uysal, Fulya AU - Berckmans, Daniel AU - Norton, Tomas AU - KU Leuven. Faculteit Bio-ingenieurswetenschappen. Opleiding Master of Bioscience Engineering. Human Health Engineering (Leuven) PY - 2019 PB - Leuven KU Leuven. Faculteit Bio-ingenieurswetenschappen DB - UniCat UR - https://www.unicat.be/uniCat?func=search&query=sysid:134741523 AB - In this thesis, we have developed a new diagnostic tool for depression and burnout by using a two step approach: first, a data-based mechanistic modelling whose results are fed in the second step into machine learning models, to classify a healthy or non healthy patient. We used ECG signals from 27 participants biking in a controlled environment as well as their power exerted during the biking as input for data-based mechanistic modelling. As a result, we could extract personalised and biologically based model features: YIC, R2, order of the A- and B- polynomials, the time delay, the slope, stressPRE, stressMIST, stressPOST, the steady state gain (SSG). These features were the outcome of the model identification phase. Furthermore, these model features were fed into a classifier by using various machine learning models, including logistic regression, k-NN, decision tree, SVM and LDA. The best accuracy for was given by decision tree classifier which is a higly interpretable model, showing how the decision making process leading to the result. Hereby, the most important model features were YIC and R2. Furthermore, its accuracy was found as 74.07% which is a promising result for a new diagnostic tool as 75% is the current state of the art. In addition, thanks to biologically-based and personalised model features, the diagnoses are believed to be more accurate and reliable. ER -