Project Overview
This project demonstrates how explainable machine learning can be applied to healthcare operations by predicting the length of hospital stays (LOS) using real-world inpatient data. Accurate LOS prediction is critical for bed management, staffing, cost control, and patient flow optimisation.
Traditional models struggle to capture complex, non-linear drivers of LOS, and many AI solutions lack transparency. This system balances performance with interpretability to support clinical trust and adoption.
Project Goal
To develop an accurate and explainable ML-based system that predicts hospital stay length while remaining interpretable and usable by healthcare professionals.
Project Results & Visual Output
The following visual outputs demonstrate the ML model’s predictive performance, feature importance for interpretability, and the interactive Streamlit application dashboard for real-time hospital stay predictions.
Hospital Length of Stay Prediction Dashboard.
Hospital Length of Stay Prediction Dashboard.
Key Features & Achievements
- Built a full ML pipeline using the SPARCS Inpatient Discharge Dataset (2021)
- Applied robust preprocessing: encoding categorical & scaling numerical features
- Trained multiple models and selected Random Forest Regressor
- Achieved average prediction error of MAE = 0.90 days
- Extracted feature importance for transparency and clinical insight
- Deployed the model as an interactive Streamlit web application
Technologies Used
Machine Learning
- Random Forest Regressor
- Decision Tree
- AdaBoost
Explainability
- Feature Importance
- XAI-Ready Architecture
Data Processing
- Pandas, NumPy
- OneHotEncoder
- StandardScaler
Evaluation & Deployment
- MAE, MSE, R² Score
- Streamlit Web App
- SPARCS Dataset (NY State)
Use Cases
- Hospital bed and capacity planning
- Patient flow and discharge optimisation
- Healthcare cost and resource management
- Clinical decision support systems
- Health analytics and operations dashboards
What This Project Demonstrates
This project demonstrates the ability to build a production-ready, explainable healthcare ML system that bridges advanced machine learning with real-world clinical usability and trust.