The doctoral dissertation in the field of Environmental Science will be examined at the Faculty of Science and Forestry, Kuopio Campus and online.
What is the topic of your doctoral research? Why is it important to study the topic?
My doctoral dissertation is about epidemiological predictive modeling, the goal of which is to enhance risk stratification by building increasingly more accurate models. The recent advances in machine learning, high-performing computing resources, and growing amount of epidemiological data have presented new opportunities for training predictive models. In practice, detecting a predisposition to certain diseases or capturing adverse changes in someone’s health status early enough helps increase the effectiveness of preventive measures and treatment. Moreover, knowing who are at higher risk allows planning preventive healthcare measures and interventions in an optimal way.
What are the key findings or observations of your doctoral research?
In this thesis, predictive models are trained using high-dimensional epidemiological data with hundreds of variables. Instead of manually preselecting known risk factors of a particular disease and running learning algorithms for models with default meta parameters, we explore high-dimensional epidemiological data with more advanced machine learning algorithms and search strategies to automatically discover informative predictors, carefully tune model meta parameters, and perform post-analysis of predictions in the end. All this not only allows building more accurate predictive models than traditional logistic regression with manual variable selection, but also applying models in a responsible way knowing how its accuracy varies for different groups of subjects.
What are the key research methods and materials used in your doctoral research?
To build predictive models on high-dimensional epidemiological data, we combine machine learning with evolutionary computation (i.e., stochastic optimization algorithms). We specifically focus on multi-objective evolutionary algorithms, which allow optimizing several objective criteria simultaneously (e.g., to maximize the model performance and minimize the model complexity). The proposed approaches have been tested on one of the most extensive datasets in the world, collected within the Kuopio Ischemic Heart Disease Risk Factor (KIHD) Study. Despite hundreds of existing studies, in which the KIHD data have been utilized, to our knowledge, the potential of KIHD as a source of high-dimensional data for predictive modeling has not been investigated yet. Based on the experimental results summarized in the thesis, we conclude that with the advanced learning procedures, high dimensionality of epidemiological data could be properly handled and turned into an advantage for predictive modeling.
The doctoral dissertation of Christina Brester, MSc, entitled Evolutionary machine learning in epidemiological predictive modeling, Examples from the Kuopio Ischemic Heart Disease Risk Factor Study, will be examined at the Faculty of Science and Forestry. The Opponent will be Professor Jonas Björk, Lund University, Sweden, and the Custos will be Professor Mikko Kolehmainen, University of Eastern Finland. Language of the public defence is English.
For more information, please contact:
Christina Brester, christina.brester@gmail.com, tel. +358 44 966 9095
- Public examination online (coming)
- Dissertation book online
