Disease Prediction Project

Choose symptoms

Type what you feel or select from the 132 dataset symptoms

0 selected

Describe symptoms:

Common symptoms:

Selected symptoms:

Loading symptom list...

Pick at least one symptom to run the model.

No prediction yet

Select symptoms on the left or type a description, then click Run prediction.

Running prediction...

Comparing selected symptoms against dataset disease profiles

Emergency warning

Call 911

Prediction results

Most likely match

Score

Symptoms matching this condition:

General precautions (from dataset):

Related symptoms to check:

Are you experiencing any of these common accompanying symptoms?

Other possible matches:

IMG

Upload an eye image

Experimental ExtraTrees model trained on 5 categories: cataracts, glaucoma, strabismus, uveitis, proptosis.

Choose an eye image (JPEG or PNG)

Click to browse or drag and drop

Upload an image to see class scores.

Extracting color and gradient statistics...

GLU

Diabetes risk model

Random Forest trained on the Pima Indians Diabetes dataset (768 records)

Fasting glucose (mg/dL)

BMI (kg/m²)

Diastolic blood pressure (mmHg)

Age (years)

2-Hour serum insulin (μU/mL)

Pregnancies (if applicable)

CVD

Stroke risk model

Trained on the Kaggle stroke prediction dataset (5,110 records)

Age

Average glucose level (mg/dL)

BMI (kg/m²)

Hypertension history

Heart disease history

Smoking status

HFR

Heart failure risk model

Trained on the Heart Failure Clinical Records dataset (299 records)

Ejection fraction (%)

Serum creatinine (mg/dL)

Age

High blood pressure history

SQL

Recent predictions

Saved locally in SQLite (predictions.db)

INFO

Model limitations

Known constraints, dataset issues, and things learned while building this project

1. Idealized symptom dataset

The disease prediction model was trained on 132 binary flags from a Kaggle dataset. Real illness presentations are nuanced, with degrees of severity and timelines that binary flags cannot capture. Predictions reflect correlation with the dataset, not clinical diagnoses.

2. Handcrafted image features

The eye classifier uses color statistics and edge gradients with an ExtraTrees model instead of a deep convolutional network. It classifies sample photos reasonably well in testing, but is easily thrown off by lighting, skin tone, or camera distance.

3. Small datasets and class imbalance

The stroke dataset has only ~4.9% positive cases, which produces false alarms even with balanced class weights. The heart failure (299 rows) and diabetes (768 rows) datasets are relatively small, which limits how well they generalize.

4. Session logging

Prediction logs are kept in a local SQLite file (predictions.db). In stateless container deployments without persistent volumes, the log resets when the container restarts.