My company produces embedded devices from which we collect a bunch of parameters (several hundred) in CSV files. Now I would like to find causal relationships between an important error parameter and all the other parameters. What I’ve done so far is training an SVM model with the important error parameter as the target and the remaining parameters as the features and then finding the features that contributed most to the target parameter. I am quite happy with the result as the most contributing labels actually make sense and the model achieves an accuracy above 90% on the test data.

But I don’t know how I could continue from these results to finding the contributing parameter in the case of a specific incident. For example, if this error parameter is triggered – how do I know, using my trained model, which parameter was the most contributing one (assuming the context of all other parameters)? At the moment the model could only tell me the likelihood of the error parameter being triggered – but not the other way around.