app.py

import streamlit as st
import pandas as pd
import joblib

# Load saved model, scaler, and expected columns
# Caching resources to improve performance
@st.cache_resource
def load_resources():
    try:
        model = joblib.load("KNN_heart.pkl")
        scaler = joblib.load("scaler.pkl")
        expected_columns = joblib.load("columns.pkl")
        
        # Load heart.csv to compute original stats for pre-scaling
        # This matches the logic in api.py and the original notebook
        df_raw = pd.read_csv("heart.csv")
        pre_scale_cols = ['RestingBP', 'Cholesterol', 'MaxHR', 'Oldpeak']
        pre_scale_stats = {}
        for col in pre_scale_cols:
            pre_scale_stats[col] = {
                'mean': df_raw[col].mean(),
                'std': df_raw[col].std()
            }
        return model, scaler, expected_columns, pre_scale_stats
    except Exception as e:
        return None, None, None, None

model, scaler, expected_columns, pre_scale_stats = load_resources()

if model is None:
    st.error("Error loading model or data files. Please ensure 'KNN_heart.pkl', 'scaler.pkl', 'columns.pkl', and 'heart.csv' are in the directory.")
    st.stop()

# Sidebar content
with st.sidebar:
    st.image("https://cdn-icons-png.flaticon.com/512/2503/2503509.png", width=100)
    st.title("About")
    st.info(
        """
        This application uses a Machine Learning model (KNN) to predict the risk of heart disease based on various health parameters.
        
        **Accuracy**: ~87% (Estimated)
        
        **Developed by**: Ankit
        """
    )
    st.markdown("---")
    st.write("### Instructions")
    st.write("1. Fill in all the details in the main form.")
    st.write("2. Click the 'Predict Risk' button.")
    st.write("3. Review the result.")

# Main content
st.title("❤️ Heart Stroke Prediction System")
st.markdown("### Please provide the patient's details below")

# Create a form to group inputs
with st.form("prediction_form"):
    col1, col2 = st.columns(2, gap="medium")

    with col1:
        st.subheader("Personal Info")
        age = st.slider("Age", 18, 100, 40, help="Age of the patient")
        sex = st.selectbox("Sex", ["M", "F"], format_func=lambda x: "Male" if x == "M" else "Female")
        resting_bp = st.number_input("Resting Blood Pressure (mm Hg)", 80, 200, 120)
        cholesterol = st.number_input("Cholesterol (mg/dL)", 100, 600, 200)
        fasting_bs = st.selectbox("Fasting Blood Sugar > 120 mg/dL", [0, 1], format_func=lambda x: "Yes" if x == 1 else "No")

    with col2:
        st.subheader("Medical History")
        chest_pain = st.selectbox("Chest Pain Type", ["ATA", "NAP", "TA", "ASY"], 
                                  help="ATA: Atypical Angina, NAP: Non-Anginal Pain, TA: Typical Angina, ASY: Asymptomatic")
        resting_ecg = st.selectbox("Resting ECG", ["Normal", "ST", "LVH"])
        max_hr = st.slider("Max Heart Rate", 60, 220, 150)
        exercise_angina = st.selectbox("Exercise-Induced Angina", ["Y", "N"], format_func=lambda x: "Yes" if x == "Y" else "No")
        oldpeak = st.slider("Oldpeak (ST Depression)", 0.0, 6.0, 1.0)
        st_slope = st.selectbox("ST Slope", ["Up", "Flat", "Down"])

    st.markdown("---")
    submitted = st.form_submit_button("Predict Heart Disease Risk")

# Logic for prediction
if submitted:
    # Create a raw input dictionary
    raw_input = {
        'Age': age,
        'RestingBP': resting_bp,
        'Cholesterol': cholesterol,
        'FastingBS': fasting_bs,
        'MaxHR': max_hr,
        'Oldpeak': oldpeak,
        'Sex_' + sex: 1,
        'ChestPainType_' + chest_pain: 1,
        'RestingECG_' + resting_ecg: 1,
        'ExerciseAngina_' + exercise_angina: 1,
        'ST_Slope_' + st_slope: 1
    }

    # Feature Engineering (High_Cholesterol)
    raw_input['High_Cholesterol'] = 1 if cholesterol > 240 else 0

    # Pre-Scaling (Manual Standardization for specific columns)
    for col in pre_scale_stats:
        if col in raw_input:
            mean = pre_scale_stats[col]['mean']
            std = pre_scale_stats[col]['std']
            # Update metric in raw_input
            raw_input[col] = (raw_input[col] - mean) / std

    # Create input dataframe
    input_df = pd.DataFrame([raw_input])

    # Fill in missing columns with 0s and reorder
    for col in expected_columns:
        if col not in input_df.columns:
            input_df[col] = 0
    input_df = input_df[expected_columns]

    # Scale the input (Final Pass)
    scaled_input = scaler.transform(input_df)

    # Make prediction
    prediction = model.predict(scaled_input)[0]

    # Show result with animation
    st.markdown("### Prediction Result")
    if prediction == 1:
        st.error("⚠️ **High Risk Detected**")
        st.warning("The model predicts a high probability of heart disease. Please consult a cardiologist immediately.")
    else:
        st.balloons()
        st.success("✅ **Low Risk Detected**")
        st.info("The model predicts a low probability of heart disease. Keep maintaining a healthy lifestyle!")