/Health-Data-Science-Competition

Primary LanguagePython

Health-Data-Science-Competition

U of SC National Big Health Data Science Competition code to classify patient chemical exposure based on their symptoms.

Basic Top-N Sweep Algorithm (achieves ~99% accuracy, top 3 conference accuracy)

  • Calculated symptom difference for each patient between all chemicals
    • Kept & sorted top-15 closest chemical symptom differences (ascending order)
  • Counted number of top-1 estimated chemical exposure per patient
  • Created top-n most likely chemicals -Only >= 5% frequency included
  • If patient predicted chemical exposure in top-n
    • Kept the same
  • If patients predicted chemical exposure not in top-n:
    • Checked ascending top-15 closest for chemical in top-n
      • If no chemical in the top-n most probable, patient symptoms are an outlier and the original chemical prediction is kept

TODO add group member algorithms TODO add neural network implementation TODO add R analysis scripts