[Splunk] Machine Larning Coordinates GPS Behavior

[oxthon]

Hello,
I ask you a question about machine learning.
I manage a small fleet of scooters that are geolocated (arduino). I want to drow their behavior.

Then, when the scooter changes, i want an alert.
So, this my first step in machine learning.
Can you help me to build it?

My data:
Date, Lat, Lon, Scooter
20190101160909 49.88890 2.32885 FBD
20190101051515 49.87281 2.33006 BVM
20190101104747 49.88349 2.33023 LMK
20190101032727 49.87316 2.32847 HJE
20190101014242 49.86916 2.32950 LXY
20190101090707 49.87989 2.33023 NWF
20190101135656 49.87558 2.33006 SXG

is it possible to detect a change in behavior of scooter?

I think i can use "Detect Numeric Outliers" of ML ToolKit.

Please, can you help me ?
Be patient with me 🙂
Thank you in advance.

OxThon

[Sukisen1981]

Each position is a category, what you need is the detect categorical outlier, here - https://docs.splunk.com/Documentation/MLApp/4.4.0/User/DCOExperiment
If you have a good historical data set this model will catch the outlier positions very effectively , give it a try

[oxthon]

Thanks you.

I don't understand ML and DCO.
How to configure Latitude and Longitude like category ?

Field to analyze
Threshold method
Écart-type
Threshold multiplier
Sliding window (# of values)
Include current point
Fields to split by
Notes

Sorry.

[oxthon]

i use the wrong api 😞

[Sukisen1981]

this is an interesting though very generic question.
do you have a data source that gets the scooter position over every 5 mins or something like that? That would have to be some sensor based API that feeds data over a rest api or the http event collector.
Now, provided you have some use case like - say at 12:00 PM scooter was at location A, then 5 minutes later it was at location B and you want to calculate outliers in the distance covered, yes your idea will work.
If a scooter is being rode too fast in terms of the distance covered over a specific time period the outlier model will work.
There can also be other use cases, like alerting based upon scooter location, that is not a ML use case though. But something like predicting usage of the scooter (total distance covered) based on historical data or the future location of a scooter based on speed (and if there is only 1 route possible) could be done.
Do you have the data/sensor for that?

[oxthon]

My case is simple for this first time.
For example, a scooter is used at one position at 12:00 pm another at 12:15 pm another at 12:30 pm ....
12:00 pm position A
12:15 position B
12:30 position C
I've just the position.
And one day, position B isn't the same.
The behavior changes.
How to model this with Splunk and ML?
I have not idea.

[Sukisen1981]

do you have the sample dataset for that?