removing of a 'three point' spike

Hi ,

I have a time series with a few spikes. A spike, in general, is a one point outlier. But as the attached image, one of the spikes in the data set is with three points. So it is not ditected by 'findpeaks'. Can anyone have an idea of how I can remove these three points.

Thank you

[KALYAN ACHARJYA]

Applying by direct threshold?

y(y>4.4)=[]

[KSSV]

Read about isoutlier https://in.mathworks.com/help/matlab/ref/isoutlier.html

[Mathieu NOE]

hello

maybe some data smoothing is also a solution to remove noise spikes

smooth, movmean, sgolayfilt

[the cyclist]

The difficult part about removing data points is never the syntax of removing them -- it is always in defining the rule that determines which points will be removed. That's where the deep thinking happens.

Often, when people here give a solution to such problems, the OP then replies, "oh but that won't work because of this other example".

So, does @KALYAN ACHARJYA's simple direct threshold solution work? It certainly removes those three points. But will it work in general? If not -- tell us why.

[Binu]

Thank you for all your ideas.

Just to make it a bit clearer, below is a section of my timeseries. I was able to find and replace all the 'single point' spikes (including inverted spikes) with linear interpolation method, as pointed by the blue arrow. But the red arrow shows the 'three point' spike which I am struggling with. I canot set a static threshold like >4.4 where it can clearout all the 'good' points.

Thank you

[Mathieu NOE]

hello again

are all those spikes usefull info or just noise ? is smoothing the right answer to the question ?

[Binu]

Hi Mathieu,

The noise is my data. I am worring only about the isolated spikes. Out of those, the '3 points' one (shown by the red arrow) that I want to get rid of. There are quite a few of those 3point spikes in my data set, so I really need to improve the script to find them rather than find and replace them one by one.

Thank you

[Mathieu NOE]

Ok

tx for the clarification

I think I have an idea : I would use one the smoothing (or enveloppe) function to create a "baseline" and then make a test condition to find all spikes above this line + slight vertical offset. If any of those point are 3 contiguous points , bingo !