This project analyzes a hotel booking dataset to identify key factors predicting cancellations and builds a machine learning model (Random Forest Classifier) to forecast potential revenue loss. The goal is to provide hotels with a data-driven tool to proactively manage cancellations and optimize revenue streams.
The main objectives of this project were to:
- Clean and preprocess the hotel booking dataset for analysis.
- Perform Exploratory Data Analysis (EDA) to identify features strongly correlated with cancellations.
- Build and train a machine learning model to predict the likelihood of a booking cancellation.
- Evaluate the model's performance using metrics like accuracy, precision, and recall.
- Provide actionable recommendations for hotels to reduce cancellations.
Hotel booking cancellations significantly impact revenue and operational efficiency. Last-minute cancellations often result in unsold rooms and lost income. This project addresses the challenge by identifying high-risk bookings, enabling hotels to implement targeted strategies like optimized overbooking, incentive offers, and proactive customer engagement to mitigate revenue loss.
The analysis used a dataset containing 119,390 booking records for a city hotel and a resort hotel, encompassing 32 features.
| Feature Category | Key Fields Used |
|---|---|
| Booking Details | hotel (type), lead_time, arrival_date details, stays_in_nights, adr |
| Guest Information | adults, children, babies, country, is_repeated_guest |
| Transaction Details | meal, market_segment, distribution_channel, deposit_type, previous_cancellations |
| Target Variable | is_canceled (1 if canceled, 0 if not) |
Data Cleaning: Preprocessing involved handling missing values (imputing company, agent, country, children), converting data types, removing invalid bookings (zero guests), feature engineering (total_stay, total_people), and removing outliers in adr (> 5000).
| Component | Tool / Language |
|---|---|
| Data Cleaning & Processing | Python (Pandas, NumPy) |
| Data Analysis & EDA | Python (Matplotlib, Seaborn) |
| Machine Learning | Python (Scikit-learn) |
| Environment | Jupyter Notebook |
| Documentation | Markdown (README), Word |
- Hotel Type Impact: City hotels experience significantly higher cancellation rates (41.7%) compared to resort hotels (27.8%).
- Lead Time: Longer lead times strongly correlate with increased cancellation likelihood.
- Price Sensitivity (ADR): Higher Average Daily Rates (ADR) are associated with more cancellations, particularly in resort hotels.
- Previous Cancellations: Guests with a history of cancellations are far more likely to cancel again.
- Deposit Type: Surprisingly, "Non-Refundable" deposits have the highest cancellation rates, potentially due to inflexibility. "No Deposit" bookings also show significant cancellations.
- Guest Loyalty: Repeated guests have a drastically lower cancellation rate.
- Model Chosen: Random Forest Classifier (selected for robustness and performance).
- Preprocessing: Categorical features were one-hot encoded. Data was split into 80% training and 20% testing sets.
- Performance Metrics:
- Accuracy: 87.2% (Correctly predicted cancellation status in over 87% of test cases).
- Precision: High precision indicated a low rate of false positives.
- Recall: Strong recall demonstrated the model's ability to identify a high percentage of actual cancellations.
Based on the analysis and model findings:
- Dynamic Deposits: Implement small, non-refundable deposits for bookings with long lead times to reduce speculative reservations.
- Reward Loyalty: Offer flexible cancellation policies or discounts to repeat guests to further secure their bookings.
- Target City Hotel Bookings: Use proactive communication (reminders, upgrade offers) for city hotel bookings due to their higher cancellation risk.
- Review Non-Refundable Policy: Consider offering future stay credits instead of complete forfeiture for non-refundable cancellations to potentially improve customer sentiment.
- Leverage Predictive Model: Integrate the model to flag high-risk bookings, enabling targeted interventions like strategic overbooking or confirmation calls.
This project successfully identified key drivers of hotel booking cancellations and developed a high-accuracy (87%) predictive model using a Random Forest Classifier. The findings and the model provide hotels with valuable tools to anticipate cancellations, manage revenue risk effectively, and implement data-driven retention strategies.