Restaurant being a customer centric business faces a lot of challenge in day-to-day operation between inventory management, staffing, and customer handling. Any mismanagement could be unhealthy in its business long-term. Medium scale restaurants often lack an appropriate sales forecasting system, and most of them still adapt to the traditional forecasting technique of intuition which comes handy with experiences.
This thesis presents a predictive analytics approach in forecasting daily sales for a restaurant setup. A real-world dataset from a restaurant in Bengaluru, India is collected to analyse the practical challenges associated. Multiple Linear Regression, KNN, Decision Tree, Random Forest and SVR are the five regression algorithms used to predict the daily sales.
Two models: Single-Output Regression and Multi-Output Regression models are trained for predicting the overall sales and individual sales of eight top-selling menu items, respectively. Influence of external factors: google trends user search pattern, hourly weather, temperature, local events are analysed.
The major contribution of this thesis would be a web application to forecast the future sales of the restaurant. The overall key factors that drive the business is identified which can be scaled and be adapted by other similar business.