I would like to predict a variable from 5 features.
I have bad scores (<50%) with all the algorithm I tried (Random Forest, Lasso, SGD, MLP).
I would like to quantitatively assess if I should give up or not on this task and learn also a thumb rule for future similar cases: I plotted the "scatter matrix"(pandas) or "pairplot"(seaborn) of my dataset and obtained mostly vertical lines.
In a 'univariate' situation I would interpret a vertical or symmetric scatter plot between feature and target as no chance of fitting it, in a multivariate situation does this still holds or could I hope in a combination of variables that could fit the target even if all features does not correlate with the target singularly?
To further investigate I also plotted the correlation matrix with very poor results too.
(1, -0.3, 0.0, 0.0, 0.0, -0.3) So my question is: are these the correct quantitative tools to assess if a multivariate fit is possible? Are there more informative test I could try? Is my task with any hope of success?