Reproducing the Paper Experiments

Contents

This folder contains all code that is necessary to replicate the experiments reported in the paper "Exploring Resampling with Neighborhood Bias on Imbalanced Regression Problems"[1].

The folder contains the following files:

• LNSmoteRegress.R the code implementing the strategies for smote for regression with neighborhood bias

• LNRandUnderRegress.R the code implementing the strategies for under-sampling with neighborhood bias

• newExpsLN.R the script for obtaining the results for the different workflows (baseline with no sampling, smoteR, random under-sampling and the proposed variants of smoteR and random under-sampling with neighborhood bias)

• Auxs.R the script with auxiliary functions for defining the workflows and the evaluation procedure

• README.md this file

Necessary Software

To replicate these experiments you will need a working installation of R. Check [https://www.r-project.org/] if you need to download and install it.

In your R installation you also need to install the following additional R packages:

• e1071
• randomForest
• earth
• performanceEstimation
• UBL
• uba

All the above packages with exception of uba, can be installed from CRAN Repository directly as any "normal" R package. Essentially you need to issue the following command within R:

install.packages(c("e1071"", "randomForest", "earth", "performanceEstimation", "UBL"))

Additionally, you will need to install uba package from a tar.gz file that you can download from [http://www.dcc.fc.up.pt/~rpribeiro/uba/].

For installing this package issue the following command within R:

install.packages("uba_0.7.7.tar.gz",repos=NULL,dependencies=T)

To replicate the figures in this repository you will also need to install the package:

• ggplot2

As with any R package, we only need to issue the following command:

install.packages("ggplot2")

Running the experiences:

Before running the experiments you need to load the data sets used in R. To obtain the 18 regression data sets and to see how you can load them, please check the README.md file in the Data folder. After having the necessary data sets, to run the experiments described in the paper you execute R in the folder with the code and then issue the command:

source("newExpsLN.R")

Alternatively, you may run the experiments directly from a Linux terminal (useful if you want to logout because some experiments take a long time to run):

nohup R --vanilla --quiet < newExpsLN.R &

Running a subset of the experiences:

Given that the experiments take long time to run, you may be interested in running only a partial set of experiments. To do this you must edit the newExpsLN.R file.

Let say, for instance, that you want to run all the workflows in only one data set. To do this you can change the instruction for(d in 1:18) in line 379.

You can also change the number and/or the values of the learning algorithms parameters. To achieve this edit the WFs list in lines 96 to 99.

After making all the necessary changes run the experiments as previously explained.

References

[1] Paula Branco, Luís Torgo and Rita P. Ribeiro (2017). "Exploring Resampling with Neighborhood Bias on Imbalanced Regression Problems" 18th Portuguese Conference on Artificial Intelligence, EPIA 2017 (to appear).