Hotel Booking Cancellation Prediction

Machine Learning project for hotel booking cancellation prediction using explainable AI and behavioral analysis.

---

Project Objective

The project aims to predict hotel booking cancellations and extend prediction into:

• behavioral interpretation
• explainable AI
• risk segmentation
• business decision support

---

Project Structure

01_Data_Processing.ipynb
│
├── Data cleaning
├── Missing value handling
├── Feature engineering
├── Leakage handling
└── Export cleaned datasets

02_Literature.ipynb
│
├── Literature summary
├── Related work
└── Project framing

03_EDA_and_Hypothesis.ipynb
│
├── Hypothesis-driven EDA
├── Visualization
├── Interaction analysis
└── Business interpretation

04_Model_Analysis.ipynb
│
├── Train/test split
├── Logistic Regression
├── Decision Tree
├── Random Forest
├── XGBoost
├── Hyperparameter tuning
└── Model evaluation

05_SHAP_Analysis.ipynb
│
├── SHAP feature importance
├── SHAP summary plots
├── Dependence plots
├── Local explanations
├── Risk segmentation
└── Behavioral personas

06_Business_Decision_Analysis.ipynb
│
├── Threshold optimization
├── Cost-sensitive evaluation
├── Calibration analysis
├── Business intervention framework
└── Final decision support analysis

---

Dataset

Dataset:

Hotel Booking Demand Dataset

Source:

https://www.kaggle.com/datasets/jessemostipak/hotel-booking-demand

Target:

is_canceled

• 1 = canceled
• 0 = not canceled

---

Required Libraries

Install required packages:

pip install pandas numpy matplotlib seaborn scikit-learn xgboost shap joblib

---

Run Order

Run notebooks in the following order:

01_Data_Processing
↓
02_Literature
↓
03_EDA_and_Hypothesis
↓
04_Model_Analysis
↓
05_SHAP_Analysis
↓
06_Business_Decision_Analysis

---

Main Outputs

Generated outputs include:

• cleaned datasets
• model-ready datasets
• EDA figures
• model evaluation results
• SHAP visualizations
• risk segmentation results
• threshold analysis
• business strategy recommendations

---

Final Model Performance

Final selected model:

Tuned XGBoost

Performance:

Metric	Value
Accuracy	0.87
Precision	0.80
Recall	0.87
F1-score	0.83
ROC-AUC	0.949

---

Notes

This repository focuses on building a complete explainable machine learning workflow rather than only maximizing prediction accuracy.