Beginning to understand Machine Learning and practical applications

BrokenArrows

I'm in the Point of Sale industry and lately I've been thinking about how i can use Machine Learning to improve speed to service and average spend value.

A simple example of what id like to achieve in a proof of concept.

Customer Orders:
1 x Double Chicken Burger
1 x Large Chips

Expected output prediction for the next item the customer might want.
1 x 12 Nuggets
1 x Large Coke

The Machine Learning model would learn from historical order history and I suppose continue to learn as new sales are made.

EDIT: In reality an item to ordered corresponds to a unique number (primary key). So really the algorithm might just be predicting what numbers are commonly grouped together.

Does anyone have a suggestion regarding what framework would suit the above scenario. Ideally .NET compatible. From my current understanding it seems there are different frameworks best suited for different tasks eg. Facial Recognition, Speach and Text Understanding, Patterns etc.

Giblet

Using Machine Learning over say, a TF-IDF based item-item recommender, might be a bit overkill, but as you want to learn...

You might be matching on a products id (as you said), attributes (burger, drink, main, dessert) or the product itself (1/2 Pounder Mega Burger with 500ml Coke), so there are things to consider when labelling your data, creating order "documents", so products in similar documents can be recommended (filtering out the existing items).

Look at using a some sort of Collaborative Filtering algorithm or even for fun, an RNN (Recurrent Neural Net) potentially with LSTM (Long Short-Term Memory) which are used when predicting sequences, especially when taking previous sequences into account.

Typical POS systems usually have a pretty short list of items (unless you are dealing with Grocery) so a standard RNN should work pretty well.

BrokenArrows

Thanks for the suggestions Giblet.

Ya its overkill and unnecessary for my example purpose, but im looking to learn and once i have an understanding i can apply it in more complex ways.

beauf

I know nothing about this. But I'm curious about it.

But I think your example isn't ideal. It's not a useful example. Perhaps a better example would be learning the effect weather and temp and seasonal patterns has on people's ordering. Thus it could learn these patterns and thus make suggestions about what stock to order in advance. Perhaps even changing advertising in advance to drive demand subconsciously.

Maybe I'm over thinking it.

L.Jenkins

POS Industry you say? Working for any of the larger companies on the west coast? On another note, would you note use machine learning for sales analytics as well? Customer behaviour on a particular day of the year and so on.

BrokenArrows

L.Jenkins wrote: »

POS Industry you say? Working for any of the larger companies on the west coast? On another note, would you note use machine learning for sales analytics as well? Customer behaviour on a particular day of the year and so on.

Yes, up until recently. But working for a customer now.

We dont use machine learning for sales analytics, but depending on how smaller scale proof of concepts pan out looking into sales analytics might be the next logical step.