app.py

import streamlit as st
import joblib
import pandas as pd
import numpy as np
import os
import time

# --- Constants and Configuration ---

MODEL_FILE = 'hotel_cancellation_prediction_model_v1_0.joblib'

# The exact list of features (columns) the model expects in this specific order
# Corrected order based on the debugging output and X_train columns
EXPECTED_FEATURES = [
    'lead_time',
    'no_of_special_requests',
    'avg_price_per_room',
    'no_of_adults',
    'no_of_weekend_nights',
    'required_car_parking_space', # Corrected position
    'no_of_week_nights',          # Corrected position
    'arrival_month',              # Corrected position
    'market_segment_type_Online'  # Corrected position
]


# Define typical ranges based on EDA (adjust as needed)
# From data.describe().T:
# lead_time: min=0, max=443, mean=81.61, 75%=118
# no_of_special_requests: min=0, max=5, mean=0.56, 75%=1
# avg_price_per_room: min=0, max=540, mean=101.03, 75%=119
# no_of_adults: min=0, max=4, mean=1.83, 75%=2
# no_of_weekend_nights: min=0, max=6, mean=0.78, 75%=1
# no_of_week_nights: min=0, max=17, mean=2.17, 75%=3
# required_car_parking_space: min=0, max=1
# arrival_month: min=1, max=12 (based on code)

FEATURE_RANGES = {
    'lead_time': {'min': 0, 'max': 450, 'default': 82},
    'no_of_special_requests': {'min': 0, 'max': 5, 'default': 1},
    'avg_price_per_room': {'min': 0.0, 'max': 600.0, 'default': 101.0},
    'no_of_adults': {'min': 0, 'max': 4, 'default': 2},
    'no_of_weekend_nights': {'min': 0, 'max': 7, 'default': 1}, # Adjusted max slightly
    'no_of_week_nights': {'min': 0, 'max': 20, 'default': 2}, # Adjusted max slightly
    'arrival_month': {'min': 1, 'max': 12, 'default': 7}, # Default to July based on EDA for example
}


# --- Model Loading (Cached) ---

@st.cache_resource
def load_cancellation_model():
    try:
        model = joblib.load(MODEL_FILE)
        return model
    except Exception as e:
        st.error(f"Error loading model: {e}")
        return None

cancellation_predictor = load_cancellation_model()

# --- Prediction Function (Critical Data Preprocessing) ---

def run_prediction(
    lead_time, market_segment_type, avg_price_per_room, no_of_adults,
    no_of_weekend_nights, no_of_week_nights, no_of_special_requests,
    arrival_month, required_car_parking_space, model
):
    """Processes inputs and runs prediction, mimicking the training feature engineering."""

    # 1. Manually construct the input row, using the required feature engineering
    data_row = {
        'lead_time': lead_time,
        'no_of_special_requests': no_of_special_requests,
        'avg_price_per_room': avg_price_per_room,
        'no_of_adults': no_of_adults,
        'no_of_weekend_nights': no_of_weekend_nights,
        'no_of_week_nights': no_of_week_nights,
        'arrival_month': arrival_month,

        # One-Hot Encoding for 'market_segment_type' (assuming 'Offline' is the base category)
        'market_segment_type_Online': 1.0 if market_segment_type == 'Online' else 0.0,

        # Binary Encoding for 'required_car_parking_space'
        'required_car_parking_space': 1.0 if required_car_parking_space == "Yes" else 0.0,
    }

    # 2. Convert dictionary to DataFrame with the correct EXPECTED_FEATURES order and dtypes
    # Explicitly set the order of columns to match EXPECTED_FEATURES
    input_df = pd.DataFrame([data_row], columns=EXPECTED_FEATURES)

    # Ensure 'required_car_parking_space' is float64 as expected by the model
    input_df['required_car_parking_space'] = input_df['required_car_parking_space'].astype('float64')


    # 3. Make Prediction
    prediction = model.predict(input_df)[0]
    # Probabilities are [Prob_Not_Canceled (0), Prob_Canceled (1)]
    probabilities = model.predict_proba(input_df)[0]
    prob_cancellation = probabilities[1]
    prob_kept = probabilities[0]


    return prediction, prob_cancellation, prob_kept


# --- Streamlit UI ---

st.set_page_config(
    page_title="Hotel Cancellation Predictor",
    layout="centered",
    initial_sidebar_state="expanded"
)

st.title("🛎️ INN Hotels: Booking Cancellation Predictor")
st.markdown("Use the controls below to input booking details and predict the cancellation risk.")

if cancellation_predictor is None:
    st.warning("Application stopped due to critical error in model loading.")
    st.stop()

# --- Input Fields (arranged for better dashboard look) ---

col1, col2 = st.columns(2)

with col1:
    lead_time = st.slider("1. Lead Time (Days before arrival)",
                          min_value=FEATURE_RANGES['lead_time']['min'],
                          max_value=FEATURE_RANGES['lead_time']['max'],
                          value=FEATURE_RANGES['lead_time']['default'])
    no_of_adults = st.number_input("4. Number of Adults",
                                   min_value=FEATURE_RANGES['no_of_adults']['min'],
                                   max_value=FEATURE_RANGES['no_of_adults']['max'],
                                   value=FEATURE_RANGES['no_of_adults']['default'],
                                   step=1)
    no_of_week_nights = st.slider("6. Number of Week Nights",
                                  min_value=FEATURE_RANGES['no_of_week_nights']['min'],
                                  max_value=FEATURE_RANGES['no_of_week_nights']['max'],
                                  value=FEATURE_RANGES['no_of_week_nights']['default'])
    arrival_month = st.selectbox("8. Arrival Month (1=Jan to 12=Dec)",
                                 list(range(FEATURE_RANGES['arrival_month']['min'], FEATURE_RANGES['arrival_month']['max'] + 1)),
                                 index=FEATURE_RANGES['arrival_month']['default'] - 1) # Adjust index for 0-based list


with col2:
    market_segment_type = st.selectbox("2. Market Segment Type", ["Online", "Offline"], index=0)
    avg_price_per_room = st.number_input("3. Average Price per Room ($)",
                                         min_value=FEATURE_RANGES['avg_price_per_room']['min'],
                                         max_value=FEATURE_RANGES['avg_price_per_room']['max'],
                                         value=FEATURE_RANGES['avg_price_per_room']['default'],
                                         format="%.2f")
    no_of_weekend_nights = st.slider("5. Number of Weekend Nights",
                                     min_value=FEATURE_RANGES['no_of_weekend_nights']['min'],
                                     max_value=FEATURE_RANGES['no_of_weekend_nights']['max'],
                                     value=FEATURE_RANGES['no_of_weekend_nights']['default'])
    no_of_special_requests = st.number_input("7. Number of Special Requests",
                                             min_value=FEATURE_RANGES['no_of_special_requests']['min'],
                                             max_value=FEATURE_RANGES['no_of_special_requests']['max'],
                                             value=FEATURE_RANGES['no_of_special_requests']['default'],
                                             step=1)
    required_car_parking_space = st.selectbox("9. Required Car Parking Space", ["Yes", "No"], index=1)


# --- Prediction Button ---

st.markdown("---") # Add a separator

if st.button("PREDICT CANCELLATION RISK", type="primary"):

    # Simple progress indicator for UX
    with st.spinner('Analyzing booking data...'):
        time.sleep(0.5)

        prediction, prob_cancellation, prob_kept = run_prediction(
            lead_time, market_segment_type, avg_price_per_room, no_of_adults,
            no_of_weekend_nights, no_of_week_nights, no_of_special_requests,
            arrival_month, required_car_parking_space, cancellation_predictor
        )


    st.markdown("---")
    st.subheader("Prediction Result")

    if prediction == 1:
        st.error(f"**High Risk of Cancellation:** The model predicts the booking will be **CANCELLED**.")
    else:
        st.success(f"**Low Risk:** The model predicts the booking will be **KEPT**.")

    st.markdown(f"**Likelihood of Cancellation: {prob_cancellation*100:.2f}%**")
    st.markdown(f"Likelihood of Keeping Booking: {prob_kept*100:.2f}%")

    if prediction == 1 and prob_cancellation > 0.70:
        st.info("💡 **Actionable Insight:** Consider proactively contacting this guest or flagging the room for immediate re-marketing.")