A Project Example: Epileptic Seizure Predictions

Epileptic Seizure Predictions Based on an Encephalogram

A Project Example

Using the machine learning techniques to predict seizures in patients based on encephalogram data.

Industry: Medical Care. Data science.

Customer: American Epilepsy Society

Project description:

This solution was initially developed for the American Epilepsy Society during the international Machine Learning development contest. The ProWide Labs solution is in the top 20 best development list among 500 teams.

It was required to develop an algorithm that would predict an epileptic seizure based on intracranial electroencephalography records.

The task was to distinguish 10-minute intervals, covering 1 hour before the seizure, and 10-minute intervals corresponding to normal brain activity (over 24 hours before the seizure).

There was about 100Gb of raw input data provided for the competition.

Scope of work:

Pre-processing of raw input data: remove noise, fix corrupted and missing data, extract various features.
Estimate the precision of predictions with various combinations of features and classification algorithms.
Cross-validation.

Technologies:

Python as a programming language
numpy, scipy, skikit-learn data science libraries
joblib, mpi4py / joblib and mpi4py libraries for parallel execution
h5py for parallel access to HDF5 files

Project size:

About 5 000 lines of Python code
2 months, 2-developer team