src/streamlit_app.py

import streamlit as st
import pandas as pd
import numpy as np
import joblib
import xgboost as xgb

st.set_page_config(
    page_title="Customer Satisfaction Prediction",
    page_icon="👤",
    layout="centered")

st.title("👤 Customer Satisfaction Prediction")
st.markdown("""
Enter the key financial details below to predict if a specific customer will be **Satisfied** or **Unsatisfied**.
*Note: Only the top influential features are shown for manual input. Others are set to default values.*
""")

# Model
@st.cache_resource
def load_model():
    try:
        model = joblib.load('src/xgb_model.pkl')
        return model
    except FileNotFoundError:
        st.error("Model file (xgb_model.pkl) not found.")
        return None

model = load_model()

# Features
if model:
    booster = model.get_booster()
    all_features = booster.feature_names
    
    importance_map = booster.get_score(importance_type='gain')
    sorted_features = sorted(importance_map.items(), key=lambda x: x[1], reverse=True)
    top_feature_names = [f[0] for f in sorted_features[:10]] # Top 10 features
    
    if 'var15' not in top_feature_names and 'var15' in all_features:
        top_feature_names.insert(0, 'var15')

# User Input 

st.sidebar.header("Customer Profile")
st.sidebar.write("Adjust the values below:")

user_inputs = {}

if model:
    for col in top_feature_names:
        label = col
        default_val = 0.0
        min_val = 0.0
        max_val = 1000000.0
        step = 1.0
        
        if col == 'var15':
            label = "Customer Age"
            default_val = 23.0
            min_val = 5.0
            max_val = 105.0
        elif col == 'saldo_var30':
            label = "Account Balance"
            default_val = 0.0
        elif col == 'var38':
            label = "Mortgage Value"
            default_val = 117310.97 
        
        val = st.sidebar.number_input(
            label=label,
            min_value=float(min_val) if col == 'var15' else None,
            value=float(default_val),
            step=step)
        user_inputs[col] = val

# Prediction 

col1, col2 = st.columns([1, 2])

with col1:
    st.image("https://cdn-icons-png.flaticon.com/512/1077/1077114.png", width=150)

with col2:
    st.subheader("Predict")
    predict_btn = st.button("Calculate Satisfaction Score", type="primary")

if predict_btn and model:
    input_data = {feature: 0 for feature in all_features}
    
    for key, value in user_inputs.items():
        if key in input_data:
            input_data[key] = value
            
    if 'var3' in input_data and 'var3' not in user_inputs:
        input_data['var3'] = 2
            
    df_single = pd.DataFrame([input_data])
    
    df_single = df_single[all_features]
    
    with st.spinner("Analyzing customer profile..."):
        prob = model.predict_proba(df_single)[:, 1][0]
        prediction = (prob > 0.5).astype(int)
        
    st.divider()
    
    col_res1, col_res2 = st.columns(2)
    
    col_res1.metric(
        label="Unhappiness Probability", 
        value=f"{prob:.2%}",
        delta="High Risk" if prob > 0.5 else "Low Risk",
        delta_color="inverse")
    
    if prediction == 0:
        col_res2.success("✅ Prediction: **SATISFIED** (Happy)")
        st.balloons()
    else:
        col_res2.error("⚠️ Prediction: **UNSATISFIED** (Unhappy)")
        st.warning("This customer is at high risk of churning or filing a complaint.")

    with st.expander("See raw input data used for model"):
        st.write(df_single)

elif not model:
    st.warning("Please upload the 'xgb_model.pkl' file to the Space.")