Heart Failure Prediction

Russell Peavey

Abstract

An example workflow demonstrating use of pandas for data processing and sklearn for modeling and evaluation.

Data

Row ordered anonymized clinical patient data of 4000 patients with ~1500 different diagnosis codes across provider visits from MIMIC-III

Feature Engineering

• SVMLight format of diagnosis code counts, min-maxed normalized, in patient observation window

Models

• Decision Tree
• Logistic Regression
• SVM

Evaluation

K-fold and randomized K-fold methods.

• F1 score
• Accuracy
• Precision
• Recall