Predictive Diagnostics

The inspiration for Predictive Diagnostics originates from high malpractice rates in the ER. This predictor is intended to be used by nurses who can confirm a physician's diagnosis based on factors such as lab test results, demographics, and patient history.

We correctly predicted kidney disease and diabetes cases in a three-class classification problem with 84.6% accuracy using logistic regression in a 10-fold cross-validation. We chose a C value of 0.1 to achieve the balance between the hyperplane overfitting the model even under a lower cost function and overgeneralizing the underlying pattern and thus misclassifying cases.

This predictive tool will decrease healthcare costs and co-pays, patient re-admittance, and defensive medicine practices, while significantly improving the quality of patient care by reducing instances of misdiagnoses and subsequently alleviating the need for specialist appointments. Medhacks 2017

Presentation Slides

Share this project: