Advanced Restaurant Reservation Predictive Model Creation

Status: Closed
Prize: $55
Entries Received: 6
Winner: You2sefHou2ri

Winner

Contest Brief

I am seeking a professional who can create a predictive model for my restaurant reservation system. Time is too tight ( Max 3 days )

Requirements:
- Python Code: (Jupiter notebook with inline comments)
- Documentation: Study report (PowerPoint)
- Required Models: Must use (LGBM Regression, Gradient Boosting Regression, Neural networks) , and you may add other models.

Notes:
-Data Utilization: The model should be developed using our reservation history data which includes the date and time of each reservation and the number of people per reservation.

-Time Frame Focus: The data used to inform this model spans over the past years.

-Model Purpose: The goal is to predict future reservation trends that help efficient planning, resource allocation and boost restaurant profits. So that the table plan is adapted, per day per service shift (midday & night). We need to increase chair occupation efficiency by minimizing the loss in tables that remain empty or unoccupied chairs.
Y = Y1 = count_attedees_booked = number of reservations that are not canceled
Y2 = number_of_client = number of people expected in the restaurant

Main Models to test:
- LGBM Regression
- Gradient Boosting Regression
- Neural networks
You can use any additional models but the above 3 are mandatory.

Ideal Skills and Experience: For this project, you'll need advanced statistical analysis skills and experience creating predictive models, specifically within the restaurant or hospitality industry. Proficiency in data visualization tools and software will be preferred as well.

Remarks:
- Use time series split in all models (TimeSerieSplit)
- Take into account Y = Y1, Y2
- Take into account public holidays
- Take weekends into account
- Remove data from the COVID lockdown period (March 15, 2020 - August 30, 2021)

Other variables (can they be target variables):
- Unusable chairs (reserved)
EXAMPLE: when you reserve a table capacity of 4 and only use 3 chairs, there will be one chair that cannot be used (unused =1) (calculated from date_time & served)
- From 'capacity' it is necessary to calculate = the number of chairs that can be used per service (i.e.: either in the evening or at midday)
- Moment = to see if it is the Midday or Evening service
- duration = the number of days before the service where the table is unlocked for everyone
- serving_date = date of service reserved

Model performance criteria:
Graphical representation (real vs pred) + confidence interval.
- Comparison by calculation (through a table of the performance criteria of each model, MAE, RMSE, R2, etc.) + graphically by drawing the RMSE curve for each model to choose the best through the graph.
- Interpretation of graphs, results.
- Parameter optimization.
- Choice of model (based on what criteria?)
- See which model is the fastest.