GSoC 2017 Shogun Data Project

gsoc-summary.md

GSoC 2017 Final Report: Data Project - Patient Monitoring and Decision Support using Health Data

Name: Olivier Nguyen

Mentors: Lea Goetz, Heiko Strathmann

Organization: Shogun Machine Learning Toolbox

Abstract

For GSoC2017, I intend to use the Shogun library on health data and show the usefulness of machine learning in applications that could save people's lives and benefit society. More specifically, I want to focus on analyzing health data for applications such as clinical decision support and mortality prediction. The dataset I will work with is the MIMIC database, which is comprised of information relating to patients admitted to the ICU at a large hospital. The data mainly includes demographic, administrative, and clinical data from over 45,000 critical care patients. The project will be divided into two parts: In the first part, I plan to perform data cleaning and apply various machine learning algorithms on the MIMIC dataset for mortality prediction, hospital readmission and length of stay. In the final part, I will explore more novel methods like LSTMs to exploit the time-series data. Recent research has shown good results of using deep learning on electronic health records.

Contributions

• Data project notebooks
  • MIMIC tutorial & data exploration
  • Dataset visualization
  • Basic machine learning model for predicting mortality & hospital length of stay
  • Feature selection & dimensionality reduction
  • Machine learning model with temporal and lagged features
  • Shogun showroom (all compiled in a notebook)
  • Basic neural networks
• My commits on shogun/develop
  • Initialising BinaryLabels & MulticlassLabels with probabilities scores
  • Added class output probabilities for Random Forest
  • Setting multiclass confidence scores in function instead of the constructor (FIX)
• Weekly blog posts
• Pending pull requests:
  • Recurrent neural networks
  • Added health data project sql files
  • Added health data project notebook for showroom