Add best prediction to classification output and metrics

[stevehadd]

Currently, when we produce an ensemble of classifiers, we produce a prediction for each, and combine them into a probability, but we should produce a best result. There are 2 ways to do this

• highest probability results
• the result from classifier that overall scores best

We should focus on the max probability class, then calculate the relevant metrics (recall, precision and F1) for the highest prob output and compare performance to individual classifiers.

[stevehadd]

We should update code to use the voting classifier https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.VotingClassifier.html

[stevehadd]

It looks the voting classifier above is slightly different to what I expected. You can't give it already trained classifier to calculate the vote. You have to give it the classifier objects then call fit. This means all the classifiers are being trained on the same data, which is different to what we are doing currently, which is producing an ensemble. For now we'll stick with the custom implementation and defer proper integration of the code into the scikit-learn framework to issue #55