appscanjs

a machine learning project to predict app by app's flow

introduction

A month ago, I found @pipcook/boa a project which makes node.js call the machine learning library of Python.So I tried！

How to use?

npm install appscanjs and ./node_modules/.bin/bip install dpkt ipaddress scikit-learn

Example

const Appscan = require('appscanjs');

const appscan = new Appscan();

const data = appscan.processor(pcapfilePath, timeThreshold, dataScale, appTimeLog);

appscan.fit(data.xTrain, data.yTrain);
const result = appscan.predict(data.xTest);

const ret = appscan.report(data.yTest, result);
console.log(ret);

API References

• new Appscan(classificationType)

  classificationType:

  1. 'RandomForestClassifier'
  2. 'RidgeClassifier'
  3. 'KNeighborsClassifier'

• processor(pcapFliePath, timeThreshold, dataScale, appTimeLog)

  • pcapFliePath: Your pcap file's path

  • timeThreshold: Default value is 1 (s), you can custom threshold. The smaller the threshold, the more data.

  • dataScale: Ratio of training data to test data. Default value is 0.7.

  • appTimeLog: Used to mark each flow.it's a object.

    Example:[{"label": 1, "closeTime": 1617691826.000627}, {"label": 2, "closeTime": 1617692026.001097}...]

    • label: app's label
    • closeTime: app's closeTime in pcapfile

    and if don't passed-in appTimeLog. processor function only return a features array.

  return a object {xTrain:[][],xTest:[][],yTrain:[],yTest:[]}

• fit(X,y)

  same as sklearn's fit. To train a classifier.

• predict(X, threshold)

  • X: the data to predict
  • threshold: use to sklearn'spredict_proba(), if the probability below threshold then prediction's label will set to -1.

  return a array as predict result.

• report(yTest, yPred)

  same as sklearn's classification_report()

Thank you for using and downloading