Automatic classification of takeaway food outlet cuisine type using machine (deep) learning

Area of Science:

• Computational epidemiology
• Public health surveillance
• Data science in health research

Background:

• Neighbourhood exposure to takeaway food outlets is not disaggregated by cuisine type.
• Manual classification of takeaway outlets is resource-intensive and hinders research on diet and obesity.
• A scalable method is needed to classify takeaway outlets by cuisine type.

Purpose of the Study:

• To develop an automated model for classifying takeaway food outlets by 10 major cuisine types using only business names.
• To overcome the resource challenges associated with manual classification.
• To enable large-scale analysis of food environments.

Main Methods:

• Utilized deep learning, specifically a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN).
• Trained the predictive model on 14,145 labelled takeaway outlets from an online food ordering platform.
• Validated model accuracy on 4,000 unseen labelled outlets.

Main Results:

• The model achieved a correct prediction rate of approximately 75% overall.
• Accuracy varied across different cuisine types.
• Successfully characterized nearly 55,000 takeaway food outlets in England by cuisine type, demonstrating its utility for public health surveillance.

Conclusions:

• Developed a novel, albeit imperfect, model to classify takeaway food outlets by 10 cuisine types using business names.
• The model employs innovative data science methods and is made available for broader use.
• Facilitates public health research and surveillance by enabling large-scale characterization of food environments.