Classifying Erroneous Data Sections of Time Series Using Machine Learning

As shown in the attached figure, I am trying to use machine learning to identify segments of erreneous data by looking at a timeseries of raw data. I created an output variable that classifies the data as 'good' or 'bad' based on how the raw data differs from the clean data. I tried inputting a single variable along with neighbors in time and found little success with the Classification Learner App. How might I be able to use machine learning/other methods to identify these erroneous data segments from just the raw data? I can clearly see a jump in the timeseries at the end of the bad data sections (where the sensors were cleaned), so I feel like an algorithm should also be able to pick that up at least.

Answer 1

[Dheeraj Singh]

You can use isoutlier to finding out outliers in your data. There are different methods that you can use for checking erroneous data.

[Jonathan Benoit]

Is there any way to pick up on the specific pattern shown in the diagram for the raw data? isoutlier doesn't pick up on the right points because the data is very variable naturally (more so than shown in the section of the diagram)