Recipe for Disaster

James has been working on a model to classify photos of food.

His company is building an application that will let users snap a picture of a plate at a restaurant and show them a potential recipe so they can cook it at home.

After a year of work, James model was working great. The company launched the model worldwide and started monitoring user feedback.

Unfortunately, users from an Asian country complained because the model wasn't working for them.

What is the most likely reason for the problem?

The model didn't have enough complexity to learn all the data, so it's normal to have problems with certain regions.

James needed to train the model for more time to fully capture the dataset's information.

The model is suffering from data drift.

The model is suffering from sampling bias.