# Univariate time series for multiple devices

I have about 50k devices, which either fail (0) or connect (1). I have data for the past 300 days for all of them. I'm trying to predict when a device will next fail. The devices are all more or less the same type.

So, I have a matrix like Device 1: [0, 1, 1, 0, 0, ...] Device 2: [1, 0, 0, 0, 0, ....] Device 3: [1, 0, 0, 0, 1, ....] ...

And trying to predict the next column.

So far, I've tried XGBoost / deep gated recurrent nets. I'm wondering if there's a better approach.

I think the data from all 50k devices should be pooled in some way, then the final prediction should be tailored to the characteristics of the individual device. Not sure how to do this statistically.

Thank you!

• Have you considered survival analysis? – user2974951 Feb 14 at 9:12
• No I haven't. Thank you for recommending this; are there good resources I should read online? – Hallo Feb 14 at 15:38
• Never mind, I just saw that your devices are failing and connecting multiple times. What other variables do you have? Can you show us a plot of one device for the DV through time? The sequence of fails and connects. – user2974951 Feb 15 at 12:48
• Sure. Is it OK if I copy paste it? This is one such device. 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 – Hallo Feb 18 at 19:44
• OK, so what other variables do you have that can help in the classification? Also are the connections and fails independent? – user2974951 Feb 19 at 7:48