XRCI contest held on 12th oct - 25th oct 2015 on hackerrank The aim of this challenge is to predict risk of death (mortality) in patients admitted to intensive care units (ICU) within hospitals.
Score :0.43/1.00
pos: 5/1500
Our approach in brief towards the problem was guided by 2 basic facts :-
a) The doctor knows the best
b) In general the trend of the vitals will ultimately determine the mortality rate for the majority of population
The features that were created therefore mainly encaptured the trend of the patient's vitals and the frequency with which the doctor is taking the tests. Some of the features were :-
- The parameters of the cubic fit(i.e fitting a cubic polynomial graph in vitals graph) of the vitals data(after it had been cleaned and it's moving avg. taken). The polynomial regression of the vital parameters allowed us to determine the trend of the vitals. For example it came to my notice that a constantly decreasing BP meant that the patient was likely to die so with the help of the following features the trend was captured.
- The difference of timestamps i.e. the frequency with which the patient's vital/lab test were taken. As more the frequency of tests implies more serious is the patient(the doctor knows best).
- The min, max and the std. deviation of the vitals wrt. the normal values. The moving std. deviation was a good measure to determine how much the values differed from normal value.
For classification model we used XGBoost(Xtreme Gradient Boosting) library.
Team Members: Shivin Srivastava, Gaurav Shrivastava, Hardik Malhotra
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent