Disregard (or weight down) "key influencers"" with a very small number of records
With the new key influencers visual, it's common for it to identify key influencers where the number of records described are only 1 or 2 (for example, "finish type quit is 11x more likely in Hanover, Maryland" when there was only one finish in Hanover Maryland and it was a quit). My idea would be to assign a weight based on number of records and adjust the influencers toward the overall population mean the lower below a threshold (30 records?) the number of records involved is.
I agree with this. Power BI claims to ignore factors that have very few records in their regression model but it doesn't seem to be the case. I have experience using this visual that telling me my influencer is something with only has 1 record, while saying my other factor with 100 records is not "influential".