# Activity prediction in a kitchen

Here is the scenario: There are three chefs(A- main chef, B and C- assistant) working together to prepare a diner set. The sequence of the event is as following.

1. Start: The three chefs enter the kitchen together with the ingredients.
2. Preparation: They prepare the utensils and the ingredients(e.g cleaning and cutting).
3. Appetizer is ready to be served.
4. Main dish is ready to be served
5. Dessert is ready to be served.
6. End: All the chefs leave the kitchen together.

My question: What machine-learning method can I use to predict the event at any point of time? Let say I can use sensors or image/video processing to collect data, what kind of data should I gather for the suggested method?

Edit: Sorry for the unclear question. What I have in mind now is, I put some sensors(or camera) to detect the start of the event with a timestamp at: the entrance, the washing corner, the cutting corner, the appetizer corner, the cooking stove(for main dish), the oven(for dessert), and the serving counter. So I have these sensor events:

1. Start time (chefs enter the kitchen)
2. Cleaning start time (ingredients preparation)
3. Cutting start time (ingredients preparation)
4. Appetizer start time
5. Stove start time (for main dish)
6. Oven start time (for dessert)
7. Serving counter (the time when one of the menu is ready to be served)
8. End time (chefs leave the kitchen)

I collect the dataset from 20 cooking cases. Applying learning algorithm to do event prediction. For example, there's a cooking started at 8:15am, cutting started at 8:45am, based on the learned data, I want to predict when will the appetizer be ready.

• Welcome to Computer Science! This is open-ended and probably has no satisfying answer, in particular because your model is ... well, saying it is "rough" would be a compliment. At this point, the best solution is probably to put a camera in the room and pay somebody to watch.
– Raphael
Jul 31 '14 at 17:06
• I would like to gather a dataset for 10 cases (or 50), apply learning algorithm to the dataset. The purpose is to do event prediction on an ongoing cooking in real time. Aug 1 '14 at 3:41
• I understand that. The problem is that you need a model.
– Raphael
Aug 1 '14 at 6:45
• @Raphael I disagree with your comment in principle: designing a model is part of science (you could say that theory is working inside a model, and application is designing and testing models). I do find the original question too vague, but I think the edit clarifies that. Aug 1 '14 at 17:11
• "What kind of data should I gather?" doesn't sound like a question about computer science to me; that sounds like something where domain knowledge about sensors and about cooking will be more useful. As far as "what machine learning algorithm can I use?", what research have you done? There is lots of information out there on available machine learning algorithms. I'd expect you to do more research before asking; we're not here to regurgitate standard textbook material. Do look at SVMs, random forests, linear regression, neural networks, k-nearest neighbors, and other standard techniques.
– D.W.
Aug 3 '14 at 6:21