Represent outlier days

4

I have hourly power consumption data for 30 days. On representing, each day data using a separate line, I get a plot as

enter image description here

I want to highlight the days with abnormally high consumption (in other words, the outlier days). I think that the current plot is too much congested. Is there any other better representation to show the outlier days?

Haroon Rashid

Posted 2016-07-16T09:44:06.663

Reputation: 290

I would like to comment on the usage of the word "outlier", and the consequent interpretation of the data: the accepted answer below fully clarifies what is the best way to visualize peak usage. However, if you want to define the outlier as anomaly, then you have to put an assumption on what the underlying statistical distribution is. In this case, if the power consumption comes for example from a powerlaw distribution, your peaks could be absolutely normal (i.e., likely to occurr). – famargar – 2017-05-05T13:15:34.237

Answers

4

One idea would be to plot the daily average power consumption in a bar plot:

enter image description here

For a finer visualization of day-hour peaks, you can try to plot it in a matrix format:

enter image description here

mc07

Posted 2016-07-16T09:44:06.663

Reputation: 76

1

Another approach would be to explore weekly or fortnightly patterns, as it is possible that your outliers may simply depend on such trends.

Another way could be also to plot on the x axis the day of the month, on the y axis the month of the year, and use different shades or colors to highlight the absolute or relative power consumption.

Joe_74

Posted 2016-07-16T09:44:06.663

Reputation: 111