CODEWITHSUNDEEP
time-serieskerassequencelstm
xlax
xlax

Reputation: 2741

LSTM for sequence prediction given a feature

I have to predict daily footfall in the carnival using the historical daily sequence of the footfall given the theme of the park on that day. The data is shown below-

I want to implement many to many LSTM to predict footfall for 9,10,11 given the themes for these days. The above table is just for the understanding of the data and the problem.

It will be really helpful if you can give me the approach to this problem. Thanks.enter image description here

Upvotes: 0

Views: 163

Answers (1)

Alex
Alex

Reputation: 1712

To my understanding an LSTM is simply not suited for this task. Basically an LSTM is pretty good for any kind of sequences, but as far as I understand your problem the footfall is only dependent on the theme. It does not matter what themes appeared on the previous days, thus it is not a sequence because a certain theme results in a certain footfall. Furthermore you have way to few data for a neural network... Anyway.

The most naive approach would be to calculate the average for your three classes c, x and y. That is the expectation for the days 9, 10 and 11. The expectations are:

c = (24 + 15 + 33) / 3 = 24
x = (20 + 32 + 17) / 3 = 23
y = (13 + 22) / 2 = 17,5

If you really want to use an neural network do the following:
1. Convert your three classes to one-hot representations (those are your classes)
e.g. c = (0 0 1), y = (0 1 0), x = (1 0 0); that is your input
2. Convert your numbers into a binary format. that is your label
3. Run the network

Upvotes: 1

Related Questions