# LINEAR LOGISITC REGRESSION

Hello, I need to simulate the training of an artificial intelligence algorithm using the following dataset: I have inspections spanning 22 years, 12 columns (representing 12 failure reasons) with outcome 0 = operational outcome and 1 = failure outcome. I need to calculate the number of days remaining before a failure occurs. Any advice?
I’m sharing the completed workflow with you. Thanks, everyone.
KNIME_.knwf (118.5 KB)

Your probabilities don’t equal 1. Could you please explain?

The system I am studying can have 12 different failure modes, so each Random Labeler node represents a failure mode in the system. I used 12 different failure probabilities. I need to calculate the time elapsed since the system state changed from operational to non-operational. System inspections are carried out once a week.

Each of the inspection probabilities > 1.

You have two different descriptions of outcome:
I need to calculate the number of days remaining before a failure occurs.
I need to calculate the time elapsed since the system state changed from operational to non-operational.

Exactly, that’s the goal; then I have to simulate additional inspections to train the linear regression algorithm.

We’re not communicating. Why do you have probabilities for each inspection that are greater than 1? Which of the two outcomes do you want? They’re different. Although I don’t agree with your probabilities, I’ve extended your workflow to calculate a classic MTBF. Will that help you?

I’m sorry for the error. The total probability is = 1. So 0.05 (1) and 0.95 (0). I would like to calculate:
The number of days remaining before a failure occurs.
The time elapsed since the system state changed from operational to non-operational.

Are the probabilities for all the inspections 0.05(1) and 0.95(0)?

The probabilities for the first Random Label Assigner are: 0.05 (1) and 0.95 (0) for all inspections. For the second Random Label Assigner node, the probabilities are: 0.004 (1) and 0.996 (0)…
KNIME_1.knwf (122.6 KB)

You need to put a little effort into this. Check ALL of the Random Label Assigners and correct them.

Hello,I’m sorry. I tried to load the updated workflow, but I’m having issues with the nodes. I performed a reset on the Random Label Assigner, but some of them are not updating.Here is the updated version
KNIME_2.knwf (122.4 KB)

Here’s a workflow that calculates MTBF and projects next predicted failure(s) as well as the distribution of times between failures. I think its “correct.” Regardless I’d be very cautious about using it in the real world.

1 Like

The axes in the histogram were mislabeled. The correction is in the hub - same link.

Thank you so much. The workflow is very interesting.

“Very interesting” is not useful feedback. I spent considerable time constructing the workflow and would like to know in some detail what did or didn’t work. More importantly the Forum is a shared community, not a place for private consulting sessions. Other KNIMERs would like to benefit from postings.

1 Like