# Detecting household showing abnormal behavior

I am trying to figure out those household which have abnormal pattern in viewing one particular youtube channels. The data is of 12 weeks

The households may fall into one of this category and many more other

1. watched that channel for all 12 weeks for few minutes.
2. watched few weeks initially for couple of minutes and now stopped watching
3. watched rarely but when they watch they spend hours (may be abnormal behavior)
4. didn't watch in the initial weeks but suddenly started watching spending long duration (abnormal behaviour)
5. earlier watched for few minutes but now suddenly spend long hours (abnormal behavior)

There are many more such cases and amongst them I need to identify who is showing abnormal behavior

I have 1000 of such households spread across a particular region somewhat in the following format.
The data is of 12 weeks starting from mid January to first week of April 2021. Each records shows how much time a household spends watching that particular channel.

year household week  Duration_in_minutes
2021       111   w1                  20
2021       111   w2                   5
2021       111   w3                  60
2021       222   w2                   2
2021       222   w3                   1
2021       222   w4                  60
2021       333   w9                   1
2021       333  w10                   2
2021       333  w11                  10
2021       333  w12                   5

Below household may be showing abnormal behavior as they started spending lot of time in the latter part.

year household week Duration_in_minutes
2021       111   w1                   5
2021       111   w2                   5
2021       111   w3                   8
2021       111   w4                   2
2021       111   w5                   1
2021       111   w6                   6
2021       111   w7                  12
2021       111   w8                   4
2021       111   w9                 250
2021       111 w110                 300

What I did was understanding each household behavior with in and storing that information in a matric. For .e.g. checking the distribution of time spent variable in the form of boxplot to see if there is any outlier or not in his own behavior. what's the average time they spent each week and if the variance is very high or not? Are they watching the channel every week or not?

With this approach in the end I am coming up with a metric having each household information and then visually inspecting is anyone showing abnormal behavior or not which I feel may not be the right way of doing it?

Is there any way to figure out may be statistically or may be using some logic to come up with households which are showing abnormal behavior.

Eagerly looking forward to some suggestion

• Maybe you could start by looking into time series clustering? Commented Apr 20, 2021 at 17:51