I don't think there is a big difference between soft sensor and software sensor, and you can use both of them. I guess that initially there was no such a term to be used in the community, some people used "software sensor" because we use or develop software to conduct the data mining/prediction, some people used "soft sensor" simply because it can be well used to distinguish "hard sensor" (conventional physical sensor). I did a simple & quick search in Web of Science; there are about 140 papers that used "software sensor" and 750 papers that used "soft sensor" (though many papers don't use either of them). I personally suggest to use "soft sensor".

Yes agree, Soft Sensor looks more relevant and logical.

Thanks for the clarification.