nsriram78/tourism-package-prediction

Tourism Package Prediction Model

Model Description

This model predicts whether a customer will purchase a Wellness Tourism Package from "Visit with Us" travel company. It uses XGBoost classifier with a custom preprocessing pipeline to handle both numeric and categorical features.

Intended Use

Primary Use: Identify potential customers for the Wellness Tourism Package to optimize marketing outreach and improve conversion rates.

Users: Sales and marketing teams at travel companies.

Out-of-scope: This model should not be used for discriminatory purposes or decisions that could significantly impact individuals' lives beyond marketing preferences.

Training Data

• Dataset: Tourism package purchase history
• Features: 18 features including customer demographics, travel preferences, and sales interaction data
  • 12 numeric features (Age, CityTier, MonthlyIncome, etc.)
  • 6 categorical features (Gender, Occupation, Designation, etc.)
• Target: Binary classification (ProdTaken: 0 = No purchase, 1 = Purchase)
• Training Set: 3302 samples
• Test Set: 826 samples
• Class Imbalance: Handled using scale_pos_weight parameter

Model Architecture

Algorithm: XGBoost Classifier with sklearn preprocessing pipeline

Preprocessing:

• Numeric features: Passthrough (no transformation)
• Nominal categorical features: OrdinalEncoder
• Ordinal feature (Designation): OrdinalEncoder with hierarchy (Executive → Manager → Senior Manager → AVP → VP)

Best Hyperparameters:

• colsample_bylevel: 0.6
• colsample_bytree: 0.6
• learning_rate: 0.15
• max_depth: 5
• n_estimators: 250
• reg_lambda: 0.5

Classification Threshold: 0.45 (optimized for F1-score)

Performance Metrics

Training Set

• Accuracy: 0.9942
• Precision: 0.9711
• Recall: 1.0000
• F1-Score: 0.9853

Test Set

• Accuracy: 0.9286
• Precision: 0.8165
• Recall: 0.8113
• F1-Score: 0.8139

How to Use

import joblib
import pandas as pd
from huggingface_hub import hf_hub_download

# Download the model
model_path = hf_hub_download(
    repo_id="nsriram78/tourism-package-prediction",
    filename="tourism_conversion_predict_model.joblib",
    repo_type="model"
)

# Load the model
model = joblib.load(model_path)

# Prepare input data (must match training feature order)
input_data = pd.DataFrame([{
    'Age': 35,
    'CityTier': 1,
    'DurationOfPitch': 15,
    'NumberOfPersonVisiting': 3,
    'NumberOfFollowups': 3,
    'PreferredPropertyStar': 4.0,
    'NumberOfTrips': 3,
    'Passport': 1,
    'PitchSatisfactionScore': 3,
    'OwnCar': 1,
    'NumberOfChildrenVisiting': 1,
    'MonthlyIncome': 22000,
    'TypeofContact': 'Self Enquiry',
    'Occupation': 'Salaried',
    'Gender': 'Male',
    'ProductPitched': 'Basic',
    'MaritalStatus': 'Married',
    'Designation': 'Manager'
}])

# Get prediction probability
prediction_proba = model.predict_proba(input_data)[0, 1]

# Apply custom threshold
prediction = (prediction_proba >= 0.45).astype(int)

print(f"Purchase Probability: {prediction_proba:.2%}")
print(f"Prediction: {'Will Purchase' if prediction == 1 else 'Will Not Purchase'}")

Training Procedure

1. Data Preparation: 80/20 train-test split with stratification
2. Hyperparameter Tuning: GridSearchCV with 5-fold cross-validation
3. Optimization Metric: F1-Score (to balance precision and recall)
4. Experiment Tracking: MLflow for logging parameters and metrics

Limitations and Considerations

• The model is trained on historical data and may not generalize to significantly different customer populations
• Performance depends on data quality and feature completeness
• Class imbalance handled but may still affect predictions on minority class
• Custom threshold of 0.45 optimized for current dataset; may need adjustment for different use cases
• Model assumes input features are in the exact order and format as training data

Ethical Considerations

• Ensure model is used responsibly for marketing purposes only
• Regularly monitor for bias in predictions across different demographic groups
• Respect customer privacy and comply with data protection regulations
• Provide opt-out mechanisms for customers who don't wish to be contacted

Model Card Authors

Sriram Narasimhan

Model Card Contact

For questions or issues, please open an issue in the model repository.