Which neural network to choose for predicting events in a smart home?

Question

Algorithms

Hello!
Question from a teapot in neural networks. I am building a smart home. Quite a lot of logic is driven by the events "everyone left home", "came back home", "went to bed", "woke up". The actual occurrence of these events is tracked. There was an idea to make a prediction of the time of occurrence of these events with constant learning on their actual occurrence. It is assumed that the time will depend on (in order of influence): day of the week, week number (even/odd), month.
To what extent is this a task that can be solved using a neural network? If a neural network makes sense, then what types/algorithms should be used? Perhaps there are ready-made solutions that can be applied to this task. I would be grateful for any hints.

Answer 1

A multilayer perceptron for each event is quite enough for itself. But - you need an initial sample. Especially since you will be using it for a month.
Whether or not a neural network is needed depends on how the accuracy of the occurrence of events fluctuates in you, and how you need to respond to these fluctuations. For example, with a probability of 99% on weekdays I come at the same time + -half an hour. Are these half hours significant? And what error will neurosity give? And if I arrive 3 hours later one day, is it worth adjusting the total time at all?

Answer 2

Thanks for the answer, I left to read about multilayer perceptrons .. I understand correctly that there should be a neural network with input parameters by the number of dependencies (current time, day of the week, etc.) and with an output parameter - the probability (?) of an event. But what about the actual occurrence of the event?
As for accuracy, I think that these possible half-hours are not significant. In any case, the efficiency of the solution should turn out to be higher than with manual planning of events. Of course, I would like that a single deviation (came 3 hours later) did not affect the result, however, if this happens regularly (some coefficient is probably needed here), then the neural network should take this into account and rebuild ..

Answer 3

You have a classic statement of the problem for the regression tree. Why use neural networks on such a large space of input values?

Answer 4

Amazon Echo