Proposition de thèse

Description :

Titre : Applications of Data Mining to Survival Prediction on Big Healthcare Data

Conférencier : Jianfei Zhang

Résumé : This project focuses on the improvement of survival prediction on Big Healthcare (or Clinical) Data, where survival prediction plays a crucial role both in personal healthcare and healthcare-related public service. There are more and more complex data in healthcare than ever before, and their size continues to increase. The baseline predictive models, including Cox and AFT, and other improved statistics-based variants perform poorly on such Big Data, especially in the presence of patients' ties; diversity of risk contributions to prediction; nonlinear correlations between predictors and clinical response. It thus prompts us to develop more effective, time-efficient big-data-mining schemes to account for these issues. The project aims to design robust algorithms and mine patient-specific subpopulations for the sake of better discovering universal patterns across tied patients; quantify risk contributions of prognostic factors and shrink noisy predictors for reducing mis-prediction rate; optimize nonlinear hazard ratio and build a parsimonious model for higher predictability and interpretability. To achieve these objectives, three programs relevant to the project are built, devoting to a new space-decomposed learning, multi-representative projection and predictor-weighted kernelizing (or regularizing), respectively. Furthermore, the project extends to the three inevitable issues in healthcare, i.e., censoring, unknown observations and healthcare privacy-preserving. Some findings in our feasibility studies declare better performances and practicability of the planned methods in comparison to existing models.

Président rapporteur : Djemel Ziou, Professeur, Département d’informatique, UdeS

Membre du jury : Ernest Monga, Professeur, Département de mathématiques, UdeS

Membre du jury : Alain Vanasse, MD, Professeur, Département de médecine de famille et de médecine d’urgence, co-directeur de recherche

Membre du jury : Shengrui Wang, Professeur, Département d’informatique, UdeS, directeur de recherche

Toutes les personnes intéressées sont cordialement invitées.