Sentry: a customizable pipeline for vulnerability prediction

Download the execution file

Click on this link to get Sentry: https://l1nk.dev/T88sq

Considering the current limitation of the existing vulnerability prediction systems, our project will have the goal to develop an engineered ML pipeline for training, validating, and exporting a vulnerability prediction model which might potentially be employed within DevOps, with a particular focus on the usability, accessibility and quality of the product.

Following a possible version of configuration file:

---

repo: https://path/to/Github/repository
dataset: "path/to/dataset.csv"

configurations:
    - 0:
          Feature Scaling: zscore
          Data Balancing: smote
          Classifier: svm
          Validation: ttsplit
          Explanation Method: permutation
    - 1:
          Feature Scaling: minmax
          Feature Selection: kbest
          K: 9
          Data Balancing: smote
          Classifier: randomforest
          Validation: ttsplit
          Explanation Method: permutation
    - 2:
          Feature Scaling: minmax
          Feature Selection: pearsoncorrelation
          Data Balancing: oversampling
          Classifier: randomforest
          Validation: kfold
          Explanation Method: confusionmatrix

statistical test:
      - 0: 1, 0
      - 1: 1, 2

The possible values for each node are:

• Data Cleaning (not exclusive)
  • dataimputation
  • shuffling
  • duplicatesremoval
• Feature Scaling
  • zscore
  • minmax
• Feature Selection
  • kbest (require K parameter)
    • K
  • variancethreshold
  • pearsoncorrelation
• Data Balancing
  • smote
  • nearmiss
  • undersampling
  • oversampling
• Classifier
  • svm
  • randomforest
  • kneighbors
• Validation
  • ttsplit
  • kfold
  • stratifiedfold
• Explanation Method (not exclusive)
  • confusionmatrix
  • permutation
  • partialdependence