/Hotel-Booking-Big-Data

Primary LanguageJupyter Notebook

Hotel Booking Prediction

How to use

pip install -r requirements.txt

Problem Description

Hotel Cancellation Prediction with Machine Learning

  • Actionable insights for hotels
  • Predict cancellations to:
    • Offer discounts and prevent revenue loss
    • Adjust staffing and optimize costs

Project Pipeline

Exploratory Data Analysis

Cancelled vs. Not Canceled:

First Observation:

  • The majority of reservations occur in the months of July and August, while the fewest bookings are made at the start and close of the year.

Second Observation:

  • City Hotels have more monthly bookings and overall bookings than Resort Hotels.
  • Both hotels have the fewest guests during the winter.

Third Observation:

  • The number of guests from Portugal is significantly higher than the other countries.
  • Portugal, Great Britain and France account for 50% of the guests.

Fourth Observation:

  • This plot clearly shows that prices in the Resort Hotel are much higher during the summer and prices of city hotel varies less.

Fifth Observation:

  • City hotels generate higher revenues compared to resort hotels across all room types.
  • Room type A are the most profitable for both city and resort hotels.

Sixth Observation:

  • There is a great loss over the years in the revenue due to cancellations for both city and resort hotels, but it is more obvious in the city hotels significantly.

Preprocessing

Handling Missing Values:

  • Get percentage of missing values in each column.
  • Drop the columns ‘agent and company’.
  • Drop the rows

Feature Engineering

Preventing Data Leakage

  • The reservation_status and reservation_status_date columns should be dropped because they provide information about when the booking was canceled or when the customer checked out of the hotel.

Encoding the categorical columns and discretizing the numerical columns

  • Encoding Categorical Columns
  • Discretizing Numerical Columns

Feature Selection

Analyzing the correlation between the target variable and the independent features

Model Development

Unsuccessful Trials

ANOVA:

  • Assumes continuous target variables and may not provide meaningful insights when applied to binary outcomes.

  • Inappropriate for capturing the relationship between categorical predictors and binary targets, leading to ineffective feature selection.

Naïve Bayes Classifier

Prior Probabilities:

  1. Mapper:

Output: [(c0, 1), (c1, 1), (c0, 1), (c0, 1), (c1, 1)]

  1. Reducer: (by key)

Output: [(c0, 3), (c1, 2)]

Likelihood Probabilities:

  1. Mapper:

Output: [(c0, (f1, v1, 5)), (c0, (f1, v2, 10)), (c1, (f1, v1, 2))]

  1. Reducer1:

Output: [(c0, (f1, v1, 50)), (c0, (f1, v2, 45)), (c1, (f1, v1, 36))]

  1. Reducer2:

Output: [(c0, (f1, v1, 50, 100)), (c0, (f1, v2, 45, 100)), (c1, (f1, v1, 36, 80))]

  1. Reducer3:

Output: [(c0, (f1, v1, 50/100)), (c0, (f1, v2, 45/100)), (c1, (f1, v1, 36/80))]

Contributors:

  • Ahmed Emad
  • Hla Hany
  • Yomna Osama
  • Youssef Mohamed