🦟 Malaria Pre-Diagnosis Prediction App

A machine learning web application that predicts malaria strain type from patient symptoms and vitals — before lab tests are conducted.

Project Structure

malaria_project/
├── app.py               ← Streamlit web app (main entry point)
├── train_models.py      ← Full training pipeline (run this first)
├── group01.csv          ← Hospital patient dataset (317 usable records)
├── models/              ← Trained model files (generated by train_models.py)
│   ├── falciparum.pkl
│   ├── malariae.pkl
│   ├── vivax.pkl
│   └── meta.pkl
├── malariapic.jpg       ← App image asset
├── requirements.txt
└── README.md

Setup & Run

# 1. Install dependencies
pip install -r requirements.txt

# 2. Train models (only needed once, or to retrain)
python train_models.py

# 3. Launch the app
streamlit run app.py

Model Details

Three separate Random Forest Classifiers (sklearn Pipeline with StandardScaler), one per malaria strain:

Strain	Positive Cases	Accuracy	ROC-AUC	CV F1
P. falciparum	153 / 317	84%	0.86	0.68 ± 0.09
P. malariae	34 / 317	92%	0.80	0.28 ± 0.05
P. vivax	120 / 317	78%	0.82	0.74 ± 0.05

P. malariae has low CV F1 due to severe class imbalance (~11% positive). class_weight='balanced' is applied.

Features Used (30 total)

Symptoms (duration in days): Fever grade, Headache, Body ache, Vomiting, Chills, Cough, Weakness, Loose stool

Severe symptoms (binary): Abdominal discomfort, Breathlessness, Rigors, Decreased appetite, Nausea, Urinal variation, Pallor, Icterus, Cyanosis, Lymphadenopathy, Diet-mixed, Drowsy, Discomfort, Abnormal appetite, Abnormal sleep

Vitals & history: Blood pressure (encoded 0–5), Pulse rate, Respiratory rate, Temperature (°F), Diabetes, Previous malaria history, Bronchial condition

Disclaimer

This tool is for clinical decision support only. It does not replace laboratory diagnosis.