About Project

• This project aims to build a predictive model to implement as a tool via a dashboard to predict whether a customer will cancel their booking or not.
• The link to the deployed dashboard is: PP5 Hotel Hocuspocus

Agile Methodology

• GitHub Project Management was used to manage this project via its kanban board feature.

• Link to the kanban board

• Link to the completed and closed tasks/issues.

• Link to the uncompleted and open tasks/issues in the backlog, which will be implemented as future features.

CRISP-DM Workflow

• In addition to Agile methodology, the Cross Industry Standard Process for Data Mining (CRISP-DM) guidelines were observed to help with the workflow of this project. The workflow has the following six stages:
  1. Business understanding
  2. Data understanding
  3. Data preparation
  4. Modelling
  5. Evaluation
  6. Deployment

Business understanding

• Determine business requirements through extensive discussions with the client.

Data understanding

• Identify, collect, and analyse the dataset that can accomplish the business requirements.

Data preparation

• Prepare the data for modelling through cleaning and feature engineering.

Modelling

• Build and assess various models based on several different modelling techniques and hyperparameters optimisation.

Evaluation

• Evaluate which model best meets the business and what to do next whether to proceed to deployment or iterate further.

Deployment

• Develop a web application using Streamlit, taking into account business requirements and principles for good user experience. The app is deployed through Heroku cloud hosting service.

Dataset Content

• The dataset is from Kaggle.
• A fictitious business scenario was created based on the dataset to conduct conventional data analysis and develop an ML predictive model to meet business requirements.
• This dataset contains booking information for a city hotel and a resort hotel, and includes information such as when the booking was made, length of stay, the number of adults, children, and/or babies, and the number of available parking spaces, among other things. All personally identifying information has been removed from the data.
• The dataset from Kaggle originally contains 100k+ entries, but in this project the dataset has been scaled down to 8k entries to minimise the size of the model pipeline for pushing to repo. It contains the following 32 variables:

No.	Variable	Description	Units
1	hotel	Type of hotel	Resort Hotel, City Hotel
2	is_canceled	Reservation cancellation status	0 = not cancelled, 1 = cancelled
3	lead_time	Number of days between booking and arrival	0-629 days
4	arrival_date_year	Year of arrival	2015-2017
5	arrival_date_month	Month of arrival	January-December
6	arrival_date_week_number	Week number of the year for arrival	1-53
7	arrival_date_day_of_month	Day of the month of arrival	1-31
8	stays_in_weekend_nights	Number of weekend nights (Saturday and Sunday) the guest stayed or booked	0-16
9	stays_in_week_nights	Number of week nights the guest stayed or booked	0-40
10	adults	Number of adults	0-27
11	children	Number of children	0-3
12	babies	Number of babies	0-2
13	meal	Type of meal booked	BB, FB, HB, SC, Undefined
14	country	Country of origin of the guest	Countries
15	market_segment	Market segment designation	Offline TA/TO, Online TA, Groups, Direct, Corporate, Complementary, Aviation
16	distribution_channel	Booking distribution channel	TA/TO, Direct, Corporate, GDS
17	is_repeated_guest	If the guest is a repeat customer	0 = not repeated, 1 = repeated
18	previous_cancellations	Number of previous bookings that were canceled by the customer	0-26
19	previous_bookings_not_canceled	Number of previous bookings that were not canceled by the customer	0-57
20	reserved_room_type	Type of reserved room	A, D, E, G, F, B, P, C, H, L
21	assigned_room_type	Type of assigned room	'A', 'D', 'E', 'C', 'G', 'B', 'K', 'F', 'P', 'H', 'I'
22	booking_changes	Number of changes made to the booking	0-14
23	deposit_type	Type of deposit made	No Deposit, Refundable, Non Refund
24	agent	ID of the travel agent responsible for the booking	1-531
25	company	ID of the company responsible for the booking	9-534
26	days_in_waiting_list	Number of days the booking was in the waiting list	0-391
27	customer_type	Type of customer	Transient, Contract, Transient-Party, Group
28	adr	Average Daily Rate	0-392
29	required_car_parking_spaces	Number of car parking spaces required	0-2
30	total_of_special_requests	Number of special requests made	0-5
31	reservation_status	Last reservation status	Check-Out, Canceled, No-Show
32	reservation_status_date	Date of the last reservation status	Dates

Business Requirements

The client wishes to come up with operational planning for minimising cancellations, improving room occupancy, and maximising revenue based on data-driven insights. After discussion with the stakeholders, the following two business requirements were agreed upon:

• Conducting conventional data analysis, the client is interested in answering the following questions: Which months have the highest number of cancellations? What are the top 5 countries with the highest number of cancellations? Which booking channels have the highest number of cancellations? Are bookings with weekend nights stay more likely to be cancelled than those with none?

• The client is interested in determining whether or not a given booking will be cancelled.

Hypothesis and how to validate?

• We suspect majority of booking cancellations are in the summer season.
  • Plot a countplot of the number of cancellations per month in descending order.
• We suspect that there are more cancellations for bookings made through distribution partners than direct bookings.
  • Plot a countplot of the number of cancellations by distribution channel in descending order.
• We suspect that bookings with weekend nights stay are more likely to be cancelled than those with none.
  • Plot a histogram of the distribution of stays in weekend nights.

The rationale to map the business requirements to the Data Visualizations and ML tasks

• Business Requirement 1: Conventional Data Analysis and Visualization

  • We will conduct conventional data analysis on the following variables: arrival_date_month, country, distribution_channel, and stays_in_weekend_nights to answer the questions from stakeholders.
  • We will plot the above variables against is_canceled to visualise insights.

• Business Requirement 2: Classification for Predicting Cancellation

  • We want to predict if a customer will cancel their booking or not. We want to build a binary classifier.

ML Business Case

Predict Cancellation

Classification Model

• We want an ML model to predict if a customer will cancel their booking or not based on historical data, which doesn't include the following variables:
  • company (irrelevant as it's the company's ID number)
  • agent (irrelevant as it's the agent's ID number)
  • country (high cardinality and might affect fitting of model)
  • arrival_date_year (model must be able to generalise future bookings)
  • reservation_status (directly related to is_cancel and might cause data leakage)
  • reservation_status_date (same reason as above)
  • assigned_room_type (same reason as above; only set in the system when guests actually check-in)
• The target variable is categorical and contains 2-classes. We consider a classification model. It is a supervised model, a 2-class, single-label, classification model output: 0 (not cancelled), 1 (cancelled)
• Our ideal outcome is to provide the client's operational planning team with reliable insight into minimising cancellations, improving room occupancy, and thus maximising revenue.
• The model success metrics are:
  • At least 80% Recall for cancellation on train and test sets
  • Version 1 of the predictive model achieves 87% Recall for cancellation on the train set, but only 66% on the test set. This suggests that the model has overfitted as there is a considerable difference.
  • Stakehoders are made aware of this performance and future improvements, such as further extensive hyperparameters optimisation or collection of new variable to add to the dataset, will be made to fine tune the model's performance.
• Heuristics: Currently, there is no approach to predict cancellations.
• The training data to fit the model comes from Kaggle. This dataset contains about 100k+ entries, but in this project the dataset has been scaled down to 8k entries to minimise the size of the model pipeline.
  • Train data - target: is_canceled; features: all other variables, except the ones listed above.

Dashboard Design (Streamlit App User Interface)

Page 1: Quick Project Summary

• Quick project summary
  • Project Terms & Jargon
  • Describe Project Dataset
  • State Business Requirements

Page 2: Data Analysis

• List the questions from the client as per business requirement 1
• Checkbox: data inspection (display the number of rows and columns in the data, and display the first ten rows of the data)
• Display the answers to the questions as bulleted list.
• Checkboxes to show each plot that provides the answers to the questions.

Page 3: Predict Cancellation

• State business requirement 2
• Set of widgets inputs, which relates to a customer's booking, to predict cancellation.
• "Run predictive analysis" button that serves the booking data to the ML pipelines and predicts if the customer will cancel their booking or not.

Page 4: Project Hypothesis and Validation

• Before the analysis, we knew we wanted this page to describe each project hypothesis, the conclusions, and how we validated each. After the data analysis, we can report that:
  • We suspect majority of booking cancellations are in the summer season.
    • Correct. Data analysis shows that August and July are the top 2 highest cancellation by month. This is during the summer season.
  • We suspect that there are more cancellations for bookings made through distribution partners than direct bookings.
    • Correct. The data analysis conducted supports this hypothesis.
  • We suspect that bookings with weekend nights stay are more likely to be cancelled than those with none.
    • False. Data analysis shows that bookings without any weekend nights stay are cancelled more than bookings with weekend nights stay.

Page 5: ML: Predict Cancellation

• Considerations and conclusions after the pipeline is trained
• Present ML pipeline steps
• Feature importance
• Pipeline performance

Manual Testing

• Please refer to the TESTING.md file for all test-related documentation.

Unfixed Bugs

• There were no known unfixed bugs in this project.

Deployment

Heroku

• The App live link is: https://pp5-hotel-hocuspocus-10951e8f369c.herokuapp.com/
• Set the runtime.txt Python version to a Heroku-20 stack currently supported version.
• The project was deployed to Heroku using the following steps:

1. Log in to Heroku and create an App
2. At the Deploy tab, select GitHub as the deployment method.
3. Select your repository name and click Search. Once it is found, click Connect.
4. Select the branch you want to deploy, then click Deploy Branch.
5. The deployment process should happen smoothly if all deployment files are fully functional. Click now the button Open App on the top of the page to access your App.
6. If the slug size is too large then add large files not required for the app to the .slugignore file.

Main Data Analysis and Machine Learning Libraries

• Jupyter: used for its interactive web application for creating the necessary notebooks to write the code for the ML project.
• Streamlit: used to create the dashboard to put their code into a web application.
• Numpy: is a Python library used for working with arrays, for example return an array of zeros with the same shape and type as a given array.
• Pandas: used for working with data sets, for example read a comma-separated values (csv) file into DataFrame.
• Matplotlib, Seaborn, and Plotly: used for visualization of the data by generating different type of plots.
• Pandas Profiling: used to create a comprehensive Report of the dataset to help with Exploratory Data Analysis (EDA).
• ppscore: used to determine the predictive power score between two columns.
• feature-engine: used to engineer the dataset’s variables and select features for use in the machine learning model.
• scikit-learn: is a Python library for machine learning used for example split randomly the train and test sets.
• imbalanced-learn: used for handling target imbalance, e.g. SMOTE.

Credits

Content

• The dataset is from Kaggle.

Coding

• This project was based on Code Institute's Churnometer walkthrough project.
• This Kaggle notebook by Farzad Nekouei was also consulted for ideas.
• This Kaggle notebook by Nitesh Yadav was also consulted for ideas.

Acknowledgements

• I would like to thank my mentor, Rohit Sharma, for his tips and support in completing this project.