Performance of Time-Based Feature Expansion Classification Method for Predicting Disease Spread
Project Introduction:
This project investigates the use of time-based feature expansion methods to enhance the accuracy of classification models in predicting disease spread using SVM and DNN algorithms. We incorporate time-related features from previous periods to refine the predictions. The goal is to create a model capable of forecasting disease spread years in advance for effective public health interventions.
Implementation Process:
- Preprocessing: check for stationarity and normalize data.
- Time-Based Feature Expansion: add historical time steps as features.
- Data Distribution: stratified 10-fold cross-validation.
Technologies Used:
- 🐍 Python
- 📊 Pandas
- ⚙️ Scikit-learn
- 📈 Matplotlib
- 🧠 TensorFlow
- 💻 Jupyter Notebook
Results and Findings:
- Model 3a (SVM): 90% accuracy with 7 features.
- Model 3b (SVM): 83% accuracy with weather-based features.
- COVID-19: 93% accuracy using 112 features (Model 7j - SVM).
- DNN: Reached 93% accuracy using 7-month time windows.
Conclusion:
This project significantly contributes to the field of predictive modeling for public health, especially for disease forecasting and medical intervention planning. The insights derived from the application of Time-Based Feature Expansion can help public health agencies proactively prepare for disease outbreaks and optimize their prevention strategies.
