Restaurant Revenue Prediction Using Machine Learning
Restaurant Revenue Prediction Using Machine Learning
Restaurant Revenue Prediction Using Machine Learning
Abstract: Currently, making a decision about when and where to open new restaurant outlets is subjective in
nature based on personal judgement and development teams' experience. This subjective data is difficult to
extrapolate across geographies and cultures. Our supervised learning algorithm will construct complex features
using simple features such as opening date for a restaurant, city that the restaurant is in, type of the restaurant (Food
Court, Inline, Drive Thru, Mobile), Demographic data (population in any given area, age and gender distribution,
development scales), Real estate data (front facade of the location, car park availability), and points of interest
including schools, banks. Applying concepts of machine learning such as support vector machines and random
forest on these parameters, it will predict the annual revenue of a new restaurant which would help food chains to
determine the feasibility of a new outlet.
Keywords: Machine Learning; Random Forest; SVM; Restaurant; Revenue; Prediction
I. INTRODUCTION
New restaurant outlets incur huge time and capital investments to establish. When the new outlet fails to break even,
the site closes within a short time and operating losses are incurred. Finding an algorithmic model to increase the
return on investments in new restaurant sites would facilitate businesses to direct their investments in other
important business areas, like innovation, and training for new employees. The problem can be defined as: design an
automated approach to decide the task environment for new restaurant by applying concepts of Support Vector
Machines, Gaussian Naive Bayes and Random Forest on certain parameters, it will predict the annual revenue of a
new restaurant outlet which would help food chains to determine its feasibility. The primary objective of Restaurant
Revenue Prediction using Machine Learning is to help restaurants make a more informed and optimal decision about
opening new outlets. It aims to find an algorithmic model to increase the effectiveness of investments in new
restaurant sites. One of the biggest features of the proposed application is that it aims to predict the revenue of new
outlets of existing restaurant chains. Analytical prediction of data has proven more effective than by human
judgement. Further, it can allow analysis and comparison of multiple new sites. Thus human errors can be avoided
and operations can be performed faster than previous methods. Given a dataset with 37 obfuscated parameters, the
algorithm will be trained on these parameters and no more. All in all, this revenue prediction system will compute an
accurate forecast of a restaurant outlets future revenues.
The system checks if the system is running in training mode. If not, it lets the user input parametric data. The data is
checked for validation. In case of invalid data, the user is asked to reenter data. Otherwise our algorithm, computes
the annual income and displays the predicted result.
IV. IMPLEMENTATION
The proposed system takes in the value of features from the user. The data fields such as opening date of the
restaurant, restaurant type, city name, city type, number of front sides, car parking and points of interest are taken as
shown in Fig. 1 below and generates the approximated Revenue depending upon inputs provided.
Fig. 3 indicates random forest (in green, uppermost line) to predicts the annual revenue as it produced a wider range
of values whereas the predictions by Gaussian Nave Bayes (in red, middle line) and SVM (in blue, lowermost line)
generated a straight line over the first thousand test examples.
VI. CONCLUSION
Thus, the concept of prediction system for future revenues of new restaurant outlets can be developed as shown.
Random forest and SVM were used to predict the annual revenue. Using this, a reference can be provided to aid
human judgement and operational losses can be minimized for food chains.
[1]
[2]
[3]
[4]
[5]
REFERENCES
SauptikDhar, Vladimir Cherkassky, Vizualization and Interpretation of SVM Classifiers, Wiley
Interdisciplinary Reviews
Geoffrey Hinton, Sam Roweis, Stochastic Neighbor Embedding, University of Toronto
Restaurant Opportunities in India: Trends and Opportunities, http://www.hvs.com/Content/1336.pdf ,
2004
Wen-Chyuan Chiang, Jason C.H. Chen, XiaojingXu, An overview of research on revenue management :
current issues and future research, International Journal of Revenue Management, Vol. 1, 2007
Dataset : Restaurant Revenue Prediction, https://www.kaggle.com/c/restaurant-revenue-prediction