I am working the sector of computer science for agriculture research. I deal here with algorithm for crop yield prediction. However, data in agriculture is very limited. To overcome the issues of small datasets + ML, I want to set up an approach that combines ML methods (learning from data) with expert knowledge.
What do I mean: E.g. Everybody knows, if you do not water your plant, it will die. If there are 90° Celsius, the plant will just burn. This knowledge is already (partially) stored in so called "crop simulation models" designed by agronomy experts. My idea was to use these expert models to generate synthetic yield data and feed this data into the training dataset for the ML models.
For me that will somehow result in an approach of "constrained machine learning" where I want to combine both. However, does some of you have any other idea how ML and expert models could be combined or the knowledge could be injected to ML methods, except via the training dataset? What other approaches exists? Google search was not too helpful here...
I am happy to hear your suggestions!
