GFS

Application with necessary methods to repeat all the experiments from the article:

On the Interpretability of Machine Learning Models and Experimental Feature Selection in Case of Multicollinear Data

Application

Application consists of a Python file gfs.py and other necessary Python modules. Required python packages are listed in the file requirements.txt.

Menu functionality

• File:
  • Load dill: load a .pkl file containing a workspace snapshot
  • Save state: save current state of necessary variables to a .pkl file
  • Load state: load saved state of necessary variables from a .pkl file
  • Clear output: delete contents of main window (text output)
  • Exit: exit the application and close all windows
• Data:
  • Provide UCI-BCW: load raw data from the source of UCI-BCW dataset
  • Provide NSL-KDD: load raw data from the source of NSL-KDD dataset (data files should be in a subfolder with a name NSL-KDD)
  • Load last UCI-BCW: load saved state of necessary variables from a UCI-BCW.pkl file
  • Load last NSL-KDD: load saved state of necessary variables from a NSL-KDD.pkl file
• Descriptive:
  • Histogram: display histogram of the class values
  • Multicollinearity: display plot of absolute values of Spearman’s rank correlation coefficients of all distinct feature pairs for the dataset, ordered by their absolute value in descending order
  • Spearman heatmap: display plot of Spearman’s rank correlation coefficients as a heatmap (see function rfpimp.plot_corr_heatmap)
  • Feature dep. matrix: display plot of dependence of the features on other features (see function rfpimp.feature_dependence_matrix)
• Base learning:
  • Learn: train a classifier from a whole dataset
  • Importance: display feature importances obtained from the trained classifier
  • Predict: run a prediction on the trained classifier using a whole dataset
  • Permut. imp.: calculate permutation feature importance using the method rfpimp.importances using a whole dataset
  • Drop-column imp.: calculate drop-column feature importance using the method rfpimp.dropcol_importances using a whole dataset
• Greedy FS:
  • Calculate: run the greedy feature selection algorithm to obtain the reduced dataset
  • Predict: run a prediction on the trained classifier using the reduced dataset
  • Cumulative feature imp.: display accuracy depending on the features, ordered descendingly by accuracy, for each step
  • Accuracy growth: display a plot of accuracy depending on steps
  • Show growth stat.: display boxplots of accuracy depending on steps from all the experiments
  • Permut. imp.: calculate permutation feature importance using the method rfpimp.importances using the reduced dataset
  • Drop-column imp.: calculate drop-column feature importance using the method rfpimp.dropcol_importances using the reduced dataset