Disease Outbreak Risk Prediction System

A machine learning-based predictive analysis system designed to forecast disease outbreak risks by analyzing epidemiological data across Philippine regions. This project implements both baseline and advanced machine learning models to predict high-risk disease locations.

---

Project Overview

This project predicts whether disease cases will increase in the next week for specific regions and diseases. It analyzes historical disease data including:

• Region - Geographic location (NCR, IV-A, IV-B, VI, etc.)
• PHI - Public Health Information provider
• Barangay - Local administrative division
• Disease - Type of disease (Measles, Chikungunya, Typhoid, etc.)
• No. of Cases - Number of reported cases
• Morbidity Week (MW) - Week of reporting

The system aims to support public health officials in resource allocation and outbreak prevention.

---

Key Features

Target Prediction

• Binary Classification: HIGH RISK (1) or LOW RISK (0)
• Prediction Variable: Will next week have above-average cases?
• Average Threshold: 4.7 cases per location
• Dataset Size: 17,755 records analyzed

Data Characteristics

• High Risk Locations: 5,519 (31%)
• Low Risk Locations: 12,236 (69%)
• Imbalanced Dataset: Requires balanced class weights in models

---

Machine Learning Models

1. Logistic Regression Model (Baseline)

A simple binary classification model that serves as the baseline for comparison.

Features Used:

• No. of Cases
• Morbidity Week (MW)

Model Configuration:

• random_state=42 - Reproducible results
• max_iter=1000 - Convergence iterations
• class_weight='balanced' - Handles class imbalance

Why These Features?

• No. of Cases - Numeric data allows sophisticated analysis (trends, rates, comparisons)
• Morbidity Week - Captures essential seasonal disease patterns and reporting cycles

Parameters Explained:

• penalty: Controls regularization (default: 'l2' - gentle enforcement)
• C: Flexibility knob (higher = more flexible)
• solver: Mathematical optimization method
• fit_intercept: Includes baseline in the model

Performance Metrics Calculated:

• Accuracy
• Precision
• Recall
• F1-Score

2. Random Forest Classifier (Advanced Model)

An ensemble learning method using multiple decision trees for improved accuracy.

Features Used:

• All 4,553 features (after one-hot encoding)
• Region, PHI, Barangay, Disease (categorical)
• No. of Cases, MW (numerical)

One-Hot Encoding Applied To:

• Region
• PHI (Public Health Information)
• Barangay
• Disease

Why All Features for Random Forest?

• Random Forest utilizes all available features because its ensemble decision-tree architecture can handle high dimensionality and identify complex patterns
• Categorical encoding creates sparse feature space which Random Forest handles well

Model Configuration:

• n_estimators=100 - Number of trees
• random_state=42 - Reproducibility
• max_depth=None - Full tree growth
• min_samples_split=2 - Node splitting threshold
• min_samples_leaf=1 - Leaf node threshold

Advantages:

• Handles high-dimensional categorical data
• Captures complex non-linear relationships
• Provides feature importance scores
• Better performance on imbalanced datasets

---

Model Evaluation Metrics

Accuracy

• Formula: (Correct Predictions) / (Total Predictions)
• Use Case: When classes are balanced
• Caveat: Can be misleading with imbalanced data
• Syntax: accuracy = accuracy_score(y_true, y_pred)

Precision

• Formula: TP / (TP + FP)
• Meaning: Of predicted HIGH risk, how many are actually correct?
• Best For: When false alarms are costly
• Syntax: precision = precision_score(y_true, y_pred)

Recall (Sensitivity)

• Formula: TP / (TP + FN)
• Meaning: Of actual HIGH risk cases, how many did we catch?
• Best For: When missing positives is dangerous (outbreak prevention)
• Syntax: recall = recall_score(y_true, y_pred)

F1-Score

• Formula: 2 × (Precision × Recall) / (Precision + Recall)
• Use Case: Balanced metric between precision and recall
• Best For: Imbalanced datasets
• Syntax: f1 = f1_score(y_true, y_pred)

---

Project Structure

Predictive-Analysis-Project/
├── README.md                          # This file
├── Machine Learning Model.ipynb       # Main analysis notebook
├── Project Manuscript.pdf             # Detailed project report
├── Datasets/                          # Disease outbreak data
│   └── GROUP#3_2DSA2_FINAL_DATASET.csv
└── Application/                       # Application code

---

Technologies & Libraries

# Data Processing
pandas                # Data manipulation
numpy                 # Numerical computing

# Visualization
matplotlib.pyplot     # Static plots
seaborn              # Statistical visualization

# Machine Learning
sklearn.model_selection  # Train-test split, stratification
sklearn.linear_model     # LogisticRegression
sklearn.ensemble         # RandomForestClassifier
sklearn.metrics          # Accuracy, Precision, Recall, F1

# Environment
Google Colab          # Cloud execution

---

Getting Started

Prerequisites

• Python 3.7+
• Jupyter Notebook or Google Colab
• Required libraries (see below)

Installation

1. Clone the repository:

git clone https://github.com/Miko-Explorer/Predictive-Analysis-Project.git
cd Predictive-Analysis-Project

2. Install dependencies:

pip install pandas numpy scikit-learn matplotlib seaborn

3. Open the notebook:

jupyter notebook "Machine Learning Model.ipynb"

Or use Google Colab for cloud-based execution (recommended for large datasets).

---

How to Use

Step 1: Load and Preview Data

from google.colab import drive
drive.mount('/content/drive')

import pandas as pd
df = pd.read_csv("path/to/GROUP#3_2DSA2_FINAL_DATASET.csv")
print(df)  # 17,928 rows x 7 columns

Step 2: Data Preparation

import numpy as np

# Sort by Region, Disease, and Morbidity Week
df = df.sort_values(['Region', 'Disease', 'MW'])

# Create target variable: Will NEXT WEEK have high cases?
df['next_week_cases'] = df.groupby(['Region', 'Disease'])['No. of Cases'].shift(-1)

# Calculate average cases
average_cases = df['No. of Cases'].mean()  # 4.7

# Create binary target: 1 if above average, 0 otherwise
df['risk_prediction'] = np.where(df['next_week_cases'] > average_cases, 1, 0)

# Remove last week for each group
df = df.dropna(subset=['risk_prediction', 'next_week_cases'])

print(f"High risk locations: {sum(df['risk_prediction']==1)}")
print(f"Low risk locations: {sum(df['risk_prediction']==0)}")

Step 3: Choose a Model

Option A - Logistic Regression (Quick Baseline):

from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score

# Select features
log_features = ['No. of Cases', 'MW']
X_log = df[log_features]
y_log = df['risk_prediction']

# Train-test split (70-30)
X_train, X_test, y_train, y_test = train_test_split(
    X_log, y_log, test_size=0.3, random_state=42, stratify=y_log
)

# Create and train model
model = LogisticRegression(random_state=42, max_iter=1000, class_weight='balanced')
model.fit(X_train, y_train)

# Evaluate
predictions = model.predict(X_test)
print(f"Accuracy: {accuracy_score(y_test, predictions):.2%}")
print(f"Precision: {precision_score(y_test, predictions):.2%}")
print(f"Recall: {recall_score(y_test, predictions):.2%}")
print(f"F1-Score: {f1_score(y_test, predictions):.2%}")

Option B - Random Forest (Better Accuracy):

from sklearn.ensemble import RandomForestClassifier

# One-hot encode categorical variables
df_encoded = pd.get_dummies(df, columns=['Region', 'PHI', 'Barangay', 'Disease'], drop_first=True)

# Prepare features
X_rf = df_encoded.drop('risk_prediction', axis=1)
y_rf = df_encoded['risk_prediction']

# Train-test split
X_train, X_test, y_train, y_test = train_test_split(
    X_rf, y_rf, test_size=0.3, random_state=42, stratify=y_rf
)

# Create and train model
rf_model = RandomForestClassifier(n_estimators=100, random_state=42, max_depth=None)
rf_model.fit(X_train, y_train)

# Evaluate
predictions = rf_model.predict(X_test)
print(f"Accuracy: {accuracy_score(y_test, predictions):.2%}")
print(f"Precision: {precision_score(y_test, predictions):.2%}")
print(f"Recall: {recall_score(y_test, predictions):.2%}")
print(f"F1-Score: {f1_score(y_test, predictions):.2%}")

---

Expected Results

Metric	Logistic Regression	Random Forest
Accuracy	~65-70%	~75-85%
Precision	Varies	Generally Higher
Recall	Varies	Generally Higher
F1-Score	Moderate	Strong

Note: Actual results depend on data split and random state

---

Key Insights

1. Class Imbalance: Dataset is imbalanced (31% HIGH, 69% LOW)

   • Solution: Use class_weight='balanced' in Logistic Regression
   • Random Forest handles this better naturally

2. Feature Engineering: One-hot encoding converts categorical variables

   • Creates 4,553 features (high dimensionality)
   • Random Forest handles sparse features well

3. Temporal Patterns: Morbidity Week captures seasonal disease patterns

   • Essential for identifying outbreak cycles

4. Geographic Variation: Region and Barangay show location-specific risks

---

Model Justification

Why Logistic Regression?

• ✅ Simple baseline for comparison
• ✅ Interpretable coefficients
• ✅ Fast training
• ✅ Requires minimal features
• ❌ Limited to linear relationships

Why Random Forest?

• ✅ Handles non-linear patterns
• ✅ Works with high-dimensional categorical data
• ✅ Provides feature importance
• ✅ Better for imbalanced data
• ✅ No feature scaling required
• ❌ Less interpretable (black box)

---

Dataset Information

Source: Philippine Disease Outbreak Data Time Period: Multiple morbidity weeks Original Records: 17,928 Processed Records: 17,755 Diseases Tracked: Measles, Chikungunya, Typhoid, etc. Geographic Scope: All Philippine regions (NCR, I-XIII, CAR, BARMM, NIR, CARAGA)

Preprocessing Steps:

1. Sorted by Region, Disease, Morbidity Week
2. Created target variable (next week prediction)
3. Removed rows with missing future data
4. Applied one-hot encoding for categorical variables
5. Stratified train-test split (70-30)

---

Contributing

Contributions are welcome! To improve this project:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/improvement)
3. Commit your changes (git commit -am 'Add enhancement')
4. Push to the branch (git push origin feature/improvement)
5. Submit a Pull Request

Suggested Improvements

• Test additional models (SVM, Gradient Boosting, Neural Networks)
• Implement hyperparameter tuning with GridSearchCV
• Add k-fold cross-validation analysis
• Create visualizations (confusion matrices, ROC curves)
• Deploy as web service (Flask, FastAPI)
• Add SHAP explainability analysis

---

Model Parameters Reference

Logistic Regression Parameters

Parameter	Default	Recommended	Effect
penalty	'l2'	'l2'	'l1': strict, 'l2': gentle, 'elasticnet': mixed
C	1.0	1.0	Lower = stricter, Higher = flexible
solver	'lbfgs'	'lbfgs'	Algorithm choice for optimization
max_iter	100	1000	Increase if convergence fails
class_weight	None	'balanced'	'balanced' for imbalanced data
random_state	None	42	For reproducibility
fit_intercept	True	True	Include baseline in model

Random Forest Parameters

Parameter	Default	Recommended	Effect
n_estimators	100	100-500	More trees = better but slower
max_depth	None	None	Deeper = more overfitting risk
min_samples_split	2	2-5	Higher = less overfitting
min_samples_leaf	1	1-2	Higher = less overfitting
random_state	None	42	For reproducibility

---

Data Distribution

Regional Distribution:

• NCR (National Capital Region)
• Regions I through XIII
• CAR (Cordillera Administrative Region)
• BARMM (Bangsamoro Autonomous Region)
• NIR (Northern Mindanao region)
• CARAGA (Caraga region)

Disease Types:

• Measles
• Chikungunya
• Typhoid
• Other reportable diseases

Temporal Coverage:

• Morbidity Weeks 1-52 (full calendar year cycle)

---

License

This project is open source and available under the MIT License.

---

Author & Contributors

Contributors

• Benedict Caliba - UST | BS Data Science and Analytics
• Thom Daniel Yutuc - UST | BS Data Science and Analytics

Author

• Miko-Explorer
• GitHub Profile

---

Support & Questions

For questions or issues:

1. Check the Project Manuscript.pdf for detailed documentation
2. Review comments in Machine Learning Model.ipynb
3. Open an issue on GitHub
4. Contact the project maintainer

---

Learning Resources

• Scikit-learn Documentation
• Logistic Regression Explained
• Random Forest Guide
• Machine Learning Metrics
• Class Imbalance Handling

---

Last Updated: June 2026 Status: Active Development Language Composition: Jupyter Notebook (98.8%), Python (1.2%) Dataset Size: 17,755 records | Model Comparison: Baseline vs Advanced