Research assistant
- Études / Statistiques / Data Deep Learning On Mimic-III : Prediction of Mortality Within 24 Hrs
This project describes the data mining on the MIMIC-III basis. The goal is to predict death in hospital based on MIMIC III. In this project, we will follow the Knowledge Discovery in Databases (KDD) process which is :
1. Selecting and extracting a set of multivariate time series data from a millon rows database by writing SQL queries.
2. Pretreat and clean the time series into a tidy dataset by exploring the data, managing missing data (missing data rate> 50%), and suppressing noise / outliers.
3. Development of a predictive model to associate a biomedical chronological series with a severity indicator (probability of mortality) by implementing several algorithms such as the gradient boost decision tree and the k-NN (k- nearest neighbors) with the DTW (Dynamic Time Warping) algorithm.
4. Result of 30% increase of the F1 score (measurement of the accuracy of a test) compared to the medical scoring index (SAPS II).
Supervisor : Pr. Agathe Guilloux | agathe.guilloux@math.cnrs.fr