I have a decent set of data and I used basic ML techniques like KNN, Random Forest, SVM, etc to do classification. Now, I found that Random Forest is giving me classification model with the best accuracy. The confusion matrix for RF classifier on the test data is:
The problem is that my classifier is getting confused between some of the classes like "Chickpea & Mustard", "Chickpea & wheat", "Mustard & wheat", "Chickpea & Wheat", etc. and hence, give me a wrong prediction. In my knowledge, this is happening because of they have highly matched signals. Now, I want to improve my classifier and specially want to reduce the confusion between highly matched signals. What are the methods I can apply to do that? Is there is any method by which I can distinguish more discreetly between highly matched signals?? Please help me out in improving my model.