Machine Learning Classification and Clustering Models

Multiclass strategies to evaluate and compare the performance of combined over$-$sampling and under$-$sampling methods at different sampling fractions when building a machine learning model for a synthetic dataset.

Instructions:

$1.$ Generate the synthetic dataset as follows: make$\_$classification$($n$\_$samples$=1000,$ n$\_$features$=8,$ n$\_$informative$=3,$ n$\_$classes$=3,$ flip$\_$y$=0\mathrm{.}2,$ weights$=[0\mathrm{.}55,0\mathrm{.}35,0\mathrm{.}1],$ random$\_$state$=42)$

$2.$ Balance the dataset $($choose any oversampling method$)$

$3.$ Considering the previous dataset, compare the accuracy performance of at least $3$ classification algorithms when using the one$-$vs$-$one and one$-$vs$-$all strategies.

$4.$ In terms of F$1$ measure, calculate the macro and weighted average performances.

$5.$ Repeat steps $3$ and $4,$ but do not implement any resampling strategy this time. Do you notice any performance change in terms of accuracy and F$1$ metrics?

$6.$ Visualize and analyze your results