In the I-IoT, analytics are classified as descriptive, diagnostic, predictive, or prescriptive:
- Descriptive analytics analyze live data to provide a comprehensive snapshot of the current status.
- Diagnostic analytics provide information about an event due to an anomaly.
- Predictive analytics add the word when to diagnostic analytics and try to predict future events. Predictive analytics not only provide information about the event that is going to happen, but also when it will happen, the risk of impact, and the probability that it will occur.
- Prescriptive analytics provide insights about the action that should be taken.
There is some overlap between these classes. They are summarized in the following table:
|
Descriptive use cases |
Diagnostic |
Predictive |
Prescriptive |
|
What is the status? |
Why is it happening? |
What's going to happen? |
What do I have to do? |
|
Expected output |
|||
|
KPI, health of the asset, efficiency, or performance. |
Alarm and root-cause analysis. |
Recommendations and/or insights. Probability that the event will occur. |
Recommendations and/or insights. |
|
Techniques |
|||
|
Rule-based, mathematical formula. |
Mathematical formula and knowledge-based. |
Statistic, deep/machine learning, data driven, regression. |
Digital twin, physical model, deep learning, reinforcement learning. |
Generally speaking, prescriptive analytics are the most complex analytics and descriptive analytics are the simplest.
Due to the large amount of data and the correlation between measures, we need at least two levels of data-processing analytics: excursion monitoring analytics (EMAs) and advanced analytics.