This project focuses on advanced classification tasks using a comprehensive patient dataset to predict critical outcomes: "Length of Stay" and mortality forecasting ("HOSPITAL EXPIRE FLAG").
Key Objectives:
- Mortality Prediction: Utilizing K-Nearest Neighbors (KNN) and Support Vector Machines (SVM) algorithms to forecast patient mortality.
- Length of Stay Prediction: Employing Neural Networks (Multilayer Perceptron - MLP) and Ensemble methods (Stacking) to predict patient length of stay.
Methods Used:
- Mortality Forecasting: Implemented KNN and SVM models to ascertain patient mortality likelihood based on historical data features.
- Length of Stay Prediction: Leveraged Neural Networks (MLP) and Ensemble techniques (Stacking) to anticipate patient length of hospitalization.
Project Scope: The project's scope includes exploratory data analysis, feature engineering, model selection, and evaluation to achieve accurate and interpretable predictions for these critical healthcare outcomes.
Standard DataScience Libraries, Scikit-Learn, Tensorflow
The dataset comprises approximately 25,000 observations related to patient admissions at the hospital. It includes standard health indicators (e.g., respiratory rate, heart rate) along with admission dates and other relevant variables. The target variables, "Length of Stay" and "HOSPITAL EXPIRE FLAG," are available in the training data for model training.
- EDA
- Data Preprocessing
- Feture Engineering
- Model Training
- SVM/KNN
- MLP/Ensemble methods (Stacking)
- Ensemble Methods
- Cross Validation
- Conclusions