AMLPRO2
Repository for 2 project for Advanced Machine Learning course
Predicted features
randomForest
- artificial: c(476, 339, 242)
- digits: c(3976, 558, 3657, 905, 3003, 339, 2302)
mcfs
- artificial: c(242, 129, 476, 106, 339)
- digits: c(3657, 3976, 558, 512, 4196, 4272) # possibly only the first three, as there is a clear step between those in score
mRMR
- artificial: c(424, 91, 277, 405, 229)
- digits: c(2433, 482, 2093, 4607, 3229, 1833, 2381)
BIC
- artificial: c(476, 49, 425)
- digits: c() # didn't compute, because there would be at least 50, if not 150 of features chosen
ReliefFexprank
- artificial: c(242, 476, 339, 106, 129)
- digits: c(3657, 558, 4508, 4387, 2302, 3464)
Dropped algorithms
Due to how long it took to compute filtered features, we decided not to use the following algorithms: Boruta, and ensemble_fs from EFS package.