app.py

import pandas as pd
import numpy as np
import tensorflow as tf
from tensorflow.keras.models import load_model
from sklearn.preprocessing import MinMaxScaler
import gradio as gr
import joblib

# Load pre-trained model and scaler
model = load_model('diabetes_model.h5')
scaler = joblib.load('scaler.pkl')

def predict_diabetes(pregnancies, glucose, insulin, bmi, age):
    """Predict diabetes probability from input features"""
    # Create input array
    input_data = np.array([[pregnancies, glucose, insulin, bmi, age]])
    
    # Scale features
    scaled_data = scaler.transform(input_data)
    
    # Make prediction
    probability = model.predict(scaled_data, verbose=0)[0][0]
    
    # Interpret results
    status = "Diabetic" if probability >= 0.5 else "Not Diabetic"
    confidence = probability if probability >= 0.5 else 1 - probability
    
    # Create explanation
    explanation = f"""
    ### Prediction: {status}  
    Confidence: {confidence:.1%}  
    
    #### Key factors contributing to this prediction:
    - Glucose level: **{'High' if glucose > 140 else 'Normal'}** ({glucose} mg/dL)
    - BMI: **{'Obese' if bmi >= 30 else 'Overweight' if bmi >= 25 else 'Normal'}** ({bmi})
    - Age: {age} years
    - Insulin: {insulin} μU/mL
    - Pregnancies: {pregnancies}
    """
    
    # Create bar chart of feature importance
    features = ['Pregnancies', 'Glucose', 'Insulin', 'BMI', 'Age']
    importance = [0.15, 0.45, 0.10, 0.20, 0.10]  # Example weights
    
    return {
        "probability": float(probability),
        "status": status,
        "explanation": explanation,
        "importance": (features, importance)
    }

# Create Gradio interface
inputs = [
    gr.Slider(0, 15, step=1, label="Number of Pregnancies"),
    gr.Slider(50, 200, value=120, label="Glucose Level (mg/dL)"),
    gr.Slider(0, 300, value=80, label="Insulin Level (μU/mL)"),
    gr.Slider(15, 50, value=32, label="BMI (kg/m²)"),
    gr.Slider(20, 100, value=33, label="Age (years)")
]

outputs = [
    gr.Label(label="Diabetes Probability"),
    gr.Markdown(label="Explanation"),
    gr.BarPlot(x="Feature", y="Importance", label="Feature Importance")
]

title = "Diabetes Prediction App"
description = "Early detection of diabetes using machine learning. Based on research: Khanam, J.J. & Foo, S.Y. (2021)"
article = """
**About this model**:  
- Trained on Pima Indians Diabetes Dataset
- Neural Network with 88.6% accuracy
- Predicts diabetes risk using 5 key health parameters
"""

gr.Interface(
    fn=predict_diabetes,
    inputs=inputs,
    outputs=outputs,
    title=title,
    description=description,
    article=article,
    examples=[
        [0, 90, 80, 24, 25],
        [3, 150, 95, 32, 35],
        [6, 180, 150, 38, 45]
    ]
).launch()