I'd like to easily find out if the five points are getting better, worse, or staying flat. This is in regard to performance of a machine learning classifier when the classifier is provided fractions of training data. So for example the fractions might be $0$-$20$, $20$-$40$, $40$-$60$, $60$-$80$, $80$-$100$. That is, given $0$-$20\%$ of the training data, how well does the classifier perform with the assumption that more data would equal better performance. The metric for performance is root mean squared error (RMSE). For a given set of classifiers, the performance data might look like this:
| index | Model | $0$ | $1$ | $2$ | $3$ | $4$ |
|---|---|---|---|---|---|---|
| $0$ | ANN | $0.66$ | $0.64$ | $0.60$ | $0.58$ | $0.56$ |
| $1$ | RF | $0.93$ | $0.95$ | $1$ | $1.01$ | $1$ |
| $2$ | Linear Regression | $1.02$ | $1.07$ | $1.07$ | $1.1$ | $1.09$ |
| $3$ | DT | $1.09$ | $1.15$ | $1.18$ | $1.23$ | $1.21$ |
| $4$ | LSTM | $0.86$ | $0.83$ | $0.81$ | $0.81$ | $0.84$ |
At first glance it might be easy to eyeball which classifiers are getting better, staying flat, etc. I can take the slope of any of these results but what is a good number for the $X$ value (assuming performance values are $Y$)? For example, $X=[0,1,2,3,4]$ and $X=[0,0.1,0.2,0.3,0.4]$ produce different slopes. Am I overthinking this?
