This project aims to analyze patient data from 130 US hospitals over a 10-year period to:
- Identify the most effective treatments for diabetes.
- Build a predictive model to determine which patients are less likely to be re-admitted after receiving certain treatments.
- Support healthcare providers with data-driven decisions to enhance patient care and reduce hospital readmissions.
- Diabetes is a chronic condition affecting millions, leading to frequent hospitalizations.
- Hospital readmissions are costly and often indicate inadequate treatment or care planning.
- By identifying effective treatments, we can reduce readmission rates and improve patient outcomes.
- Improve Patient Care: Enable personalized and effective treatment plans.
- Reduce Costs: Minimize unnecessary hospital visits and readmissions.
- Support Clinical Decisions: Offer predictive insights to healthcare professionals.
- Real-world data from 130 hospitals, spanning 10 years.
- Features include: patient demographics, diagnoses, medications, procedures, lab results, and readmission status.
-
Data Preprocessing
- Removed duplicates, handled missing values.
- Encoded categorical variables and filtered diabetes-related cases.
-
Exploratory Data Analysis (EDA)
- Visualized patient distributions and treatment outcomes.
- Identified patterns and correlations between treatments and readmissions.
-
Feature Engineering
- Created derived features:
inpatient_visits,medication_change,length_of_stay, etc. - Labeled readmission as:
within 30 days,after 30 days, ornever.
- Created derived features:
-
Model Building
- Applied classification models:
- Logistic Regression
- Random Forest
- XGBoost
- Target variable:
readmission status.
- Applied classification models:
-
Model Evaluation
- Metrics used: Accuracy, Precision, Recall, F1-score, ROC-AUC.
- Performed cross-validation and hyperparameter tuning.
-
Insights & Recommendations
- Identified treatments that correlate with low readmission rates.
- Recommended treatment protocols based on predictive insights.
- Successfully trained a model to predict patient readmission likelihood.
- Gained actionable insights on effective treatments for diabetes.
- Demonstrated how machine learning can assist in improving healthcare systems and patient well-being.
- Languages: Python, Pandas, NumPy
- Visualization: Matplotlib, Seaborn
- Modeling: Scikit-learn, XGBoost
- Tools: Jupyter Notebook, Git
- Include more recent or real-time patient data.
- Deploy the model as a REST API for hospital decision support systems.
- Expand to other chronic diseases for treatment analysis.
Ashish Raj
GitHub: ashishraj570raj