Healthcare Patient Outcome Prediction -- 2
Budget: ₹750 – ₹1,250 INR
I need to turn historical patient-record data into a reliable model that predicts individual outcomes and flags high-risk cases early. The raw material is a sizeable set of anonymised electronic health records: demographics, vitals, lab results, admission notes, discharge summaries—typical structured tables mixed with a handful of free-text fields that have already been de-identified.
Here is what I’m after:
• A clean, reproducible pipeline in Python and SQL that pulls the records, handles missing values, encodes categorical variables, and outputs a model-ready dataset.
• One or more predictive models built in scikit-learn (feel free to benchmark XGBoost or similar) together with well-explained feature importance. AUROC is the key metric; I’d like to see cross-validation and a hold-out test set.
• Clear visualisations—Power BI dashboards are preferred—showing cohort characteristics, model performance, and an interactive patient-risk explorer that clinical staff can filter by age group, diagnosis, or ward.
• A concise report (Jupyter Notebook is fine) walking through assumptions, statistical tests, and next-step recommendations for improving patient care.
I already have secure database access plus sample CSV exports. Once you’re in, just push code to a private Git repo so the hospital IT team can review line‐by‐line. Please estimate timeline and any data-governance needs up front; HIPAA compliance is non-negotiable.
Here is what I’m after:
• A clean, reproducible pipeline in Python and SQL that pulls the records, handles missing values, encodes categorical variables, and outputs a model-ready dataset.
• One or more predictive models built in scikit-learn (feel free to benchmark XGBoost or similar) together with well-explained feature importance. AUROC is the key metric; I’d like to see cross-validation and a hold-out test set.
• Clear visualisations—Power BI dashboards are preferred—showing cohort characteristics, model performance, and an interactive patient-risk explorer that clinical staff can filter by age group, diagnosis, or ward.
• A concise report (Jupyter Notebook is fine) walking through assumptions, statistical tests, and next-step recommendations for improving patient care.
I already have secure database access plus sample CSV exports. Once you’re in, just push code to a private Git repo so the hospital IT team can review line‐by‐line. Please estimate timeline and any data-governance needs up front; HIPAA compliance is non-negotiable.