Is it possible to simulate a classifiers behaviour?

I am training a few classification models, KNN, SVM and Weighted KNN on my dataset (attached). During training, I keep getting an error rate of 0% for the weighted knn, no matter the number of neighbours. I even tried 100 neighbours, but I again got a 0% error. This doesn't seem possible to me.

So I was wondering if one can simulate the models behaviour, to visually see how the algorithm makes its decision based on any dataset, or a random set of datapoints. My goal is to also have it animated and running on a loop. I would like to first do it on the WkNN as that is what I am most concerned with.

I have tried a decision surface/boundary but that is a static image of the boundary. I was wondering if one can watch an algorithm make its decision, live.

Any help would be greatly appreciated.

Thank you.

Answer 1

[Aditya Patil]

I understand that you would like to have options for interpreting the results in real time. Currently, it's not possible to visualize these models in real time. The models such as KNN and SVM rely on a decision boundary to decide the class of a data sample. As such, visualizing the decision boundary is a good option to interpret the results.

For the specific dataset attached, the issue seems to be with high dimensionality. To get a more robust model, you can utilize one of the dimensionality reduction techniques that are available.