I have a dataset which has timestamp and goods variable for an object like the following example:

  • On Moday, John bought Beer, chicken, pizza.
  • On Tuesday, John bought pizza, pork and beef.
  • On Wednesday, John bought beer, coke and Vegetable

Can I use machine learning to predict what good will he buy in a specific time and what model can be applied?

Can we answer the question: "What will he buy on Thursday?"

  • 1
    $\begingroup$ Please tell us more. What kind of prediction do you want to make? What factors do you think will determine what purchases he makes next? Can you observe any of those factors, and if so, which? $\endgroup$ – D.W. Jul 6 '16 at 22:50
  • 1
    $\begingroup$ Virtual vote to close as "unclear" since there is not enough information here for a meaningful discussion, let alone answers. Community votes, please! $\endgroup$ – Raphael Jul 7 '16 at 0:07
  • $\begingroup$ I edited the question for more clearly. $\endgroup$ – lotusirous Jul 7 '16 at 2:45
  • $\begingroup$ I agree with @Raphael. Can the OP state what has been already attempted? $\endgroup$ – wabbit Jul 7 '16 at 3:22
  • $\begingroup$ @D.W. Can I observe the data of buying transactions to predict what goods will he buy in the future? $\endgroup$ – lotusirous Jul 7 '16 at 3:37

Short answer:

Yes it can be done using machine learning if you have features with sufficient information.

Long answer:

You can model this either as a multilabel problem (i.e Beer, chicken, pizza are multiple labels [non mutually exclusive] for the same data point) or as a multiclass problem (by creating a combination like "Beer, chicken, pizza". In the multiclass case the labels would be mutually exclusive). You can use day of the week, name of person and some other attributes as features (aka predictors) for the model.

For a more detailed answer you'd need to tell us what you have already tried out so that the community can help you. Also, see this related question.


You might be interested in techniques from data mining, especially association rule mining and frequent itemset mining. They try to detect which combinations of products tend to be bought together (e.g., if you buy pizza you'll often buy beer too).

  • $\begingroup$ Based on this dissertation. Assocication rule is tried to discover the patterns in the data. I don't think it can be applied in case of mine. $\endgroup$ – lotusirous Jul 7 '16 at 2:54

Not the answer you're looking for? Browse other questions tagged or ask your own question.