This is an abridged walkthrough of my final project at INSOFE Bangalore, first published on Kaggle. It tackles an interesting data science problem and contains many visualizations, insights and predictive modelling techniques.
We were given some data from a US hospital and the task was to predict which patients will be readmitted within 30 days of their discharge. The target classes are - readmitted within 30 days (positive class) and not readmitted within 30 days (negative class). We were asked to optimize our models for Recall or true positive rate.
This script goes through the following steps:
- Data preparation
- Exploratory Data Analysis
- Feature engineering
- Data visualizations
- Data preprocessing
- Learning Curves
- Model building and validation
- Patterns and rules extraction
- Prediction on test data
- Summary and conclusion