This project aims to predict the admission of confirmed COVID-19 cases to ICU using machine learning techniques. By early prediction, it assists in resource allocation and patient care planning, contributing to reducing mortality rates.
The unprecedented COVID-19 pandemic necessitated proactive measures to manage healthcare resources efficiently. Early ICU admission prediction aids in resource allocation and prioritization, potentially saving lives.
- Size: 5000 rows, 85 columns
- Source: Obtained as part of a data science course
- Features: Various clinical, demographic, and diagnostic indicators
-
Exploratory Data Analysis (EDA):
- Utilized Pandas profiling, Seaborn, Matplotlib, and Plotly for comprehensive insights.
- Conducted data cleaning to handle duplicates, null values, and outliers.
-
Feature Selection:
- Employed 7 algorithms including Pearson correlation, SelectFromModel, SelectKBest, Recursive Feature Elimination, and VarianceThreshold to reduce dimensionality to 35.
-
Model Building:
- Explored 13 classification algorithms such as Logistic Regression, SVM, Decision Trees, Random Forest, AdaBoost, Gradient Boosting, XGBoost, KNN, and Naive Bayes.
- Utilized GridSearchCV for hyperparameter tuning.
- Implemented a voting classifier to ensemble top-performing models.
-
Evaluation:
- Achieved a testing accuracy of 85%, precision of 91%, and recall of 84%.
- Pandas
- NumPy
- Seaborn
- Matplotlib
- Plotly
- Scikit-learn
This project demonstrates the efficacy of machine learning in predicting ICU admissions for COVID-19 patients. The insights gained can inform healthcare resource allocation strategies and aid in optimizing patient care.