Greeting.
I have been trying to train datasets for land cover change prediction using MLP, and my accuracy rate have been ranging from 11 to 13.56%, which is not even close to the recommended percentage (80%).
I understand that you can either use the default setting in the Run Transition Sub-Model panel or use insert your own values. I have tried both, but the accuracy rate is still the same. Any ideas on what I should do next?
Thanks.
Hi, If you are using the default values and getting low skill and accuracy rates, then the first place to revisit is your model, particularly, the transitions used in the model and the variables used to model those transitions historically. Most likely the variables you are using do not adequately explain the historical change. You may have to add more or alter the variables themselves. The Help system and the Manual give more detail on this. Rethink the model and try to find more explanatory variables that address your transitions.
James
Hi, I have a follow up question. I would like to know, how the accuracy of the basemaps is influencing the accuracy rates of the change prediction? Is it even possible to achieve a high accuracy rate for transitions, that are difficult to classify in the basemaps (i.e. shrubs, grasslands, irrigated or rainfed crops)?
What is the source of the recommended 80% accuracy rate ? Would a lower accuracy rate of 60% still be acceptable if multiple and complex transitions are predicted with MLP? Thanks
