--- license: mit library_name: sklearn tags: - classification - phishing --- # Model description ## Training Procedure ### Hyperparameters

Click to expand

| Hyperparameter | Value | |-------------------------------------------------------------|--------------------------| | memory | | | steps | [('standardscaler', StandardScaler()), ('calibratedclassifiercv', CalibratedClassifierCV(cv=5, estimator=RandomForestClassifier(),
method='isotonic'))] | | verbose | False | | standardscaler | StandardScaler() | | calibratedclassifiercv | CalibratedClassifierCV(cv=5, estimator=RandomForestClassifier(),
method='isotonic') | | standardscaler__copy | True | | standardscaler__with_mean | True | | standardscaler__with_std | True | | calibratedclassifiercv__base_estimator | deprecated | | calibratedclassifiercv__cv | 5 | | calibratedclassifiercv__ensemble | True | | calibratedclassifiercv__estimator__bootstrap | True | | calibratedclassifiercv__estimator__ccp_alpha | 0.0 | | calibratedclassifiercv__estimator__class_weight | | | calibratedclassifiercv__estimator__criterion | gini | | calibratedclassifiercv__estimator__max_depth | | | calibratedclassifiercv__estimator__max_features | sqrt | | calibratedclassifiercv__estimator__max_leaf_nodes | | | calibratedclassifiercv__estimator__max_samples | | | calibratedclassifiercv__estimator__min_impurity_decrease | 0.0 | | calibratedclassifiercv__estimator__min_samples_leaf | 1 | | calibratedclassifiercv__estimator__min_samples_split | 2 | | calibratedclassifiercv__estimator__min_weight_fraction_leaf | 0.0 | | calibratedclassifiercv__estimator__n_estimators | 100 | | calibratedclassifiercv__estimator__n_jobs | | | calibratedclassifiercv__estimator__oob_score | False | | calibratedclassifiercv__estimator__random_state | | | calibratedclassifiercv__estimator__verbose | 0 | | calibratedclassifiercv__estimator__warm_start | False | | calibratedclassifiercv__estimator | RandomForestClassifier() | | calibratedclassifiercv__method | isotonic | | calibratedclassifiercv__n_jobs | |

### Model Plot This is the architecture of the model loaded by joblib.

Pipeline(steps=[('standardscaler', StandardScaler()),('calibratedclassifiercv',CalibratedClassifierCV(cv=5,estimator=RandomForestClassifier(),method='isotonic'))])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

## Evaluation Results | Metric | Value | |-----------|----------| | accuracy | 0.945652 | | f1-score | 0.945114 | | precision | 0.951996 | | recall | 0.938331 | ## Test Report ![Test Report](plots/classification_report.png) ## Model Interpretation ### Feature Importances ![Test Report](plots/feature_importances.png) # How to Get Started with the Model Below are some code snippets to load the model. ## With joblib (not recommended) ```python import joblib import pandas as pd urls = [ { "url": "https://www.rga.com/about/workplace", "nb_hyperlinks": 97.0, "ratio_intHyperlinks": 0.969072165, "ratio_extHyperlinks": 0.030927835, "ratio_extRedirection": 0.0, "safe_anchor": 25.0, "domain_registration_length": 3571.0, "domain_age": 11039, "web_traffic": 178542.0, "google_index": 0.0, "page_rank": 5, }, ] model = joblib.load("models/model.pkl") df = pd.DataFrame(urls) df = df.set_index("url") probas = model.predict_proba(df.values) for url, proba in zip(urls, probas): print(f"URL: {url['url']}") print(f"Likelihood of being a phishing site: {proba[1] * 100:.2f}%") print("----") # output: # URL: https://www.rga.com/about/workplace # Likelihood of being a phishing site: 0.89% # ---- ``` ## With ONNX (recommended) ### Python ```python import numpy as np import onnxruntime import pandas as pd # Defining a list of URLs with characteristics urls = [ { "url": "https://www.rga.com/about/workplace", "nb_hyperlinks": 97.0, "ratio_intHyperlinks": 0.969072165, "ratio_extHyperlinks": 0.030927835, "ratio_extRedirection": 0.0, "safe_anchor": 25.0, "domain_registration_length": 3571.0, "domain_age": 11039, "web_traffic": 178542.0, "google_index": 0.0, "page_rank": 5, }, ] # Initializing the ONNX Runtime session with the pre-trained model sess = onnxruntime.InferenceSession( "models/model.onnx", providers=["CPUExecutionProvider"], ) # Creating a DataFrame from the list of URLs df = pd.DataFrame(urls) df = df.set_index("url") # Converting DataFrame data to a float32 NumPy array inputs = df.astype(np.float32).to_numpy() # Using the ONNX model to make predictions on the input data probas = sess.run(None, {"X": inputs})[1] # Displaying the results for url, proba in zip(urls, probas): print(proba) print(f"URL: {url['url']}") print(f"Likelihood of being a phishing site: {proba[1] * 100:.2f}%") print("----") # output: # URL: https://www.rga.com/about/workplace # Likelihood of being a phishing site: 0.89% # ---- ``` ### JavaScript ```javascript const ort = require('onnxruntime-node'); const urls = [ { "url": "http://rapidpaws.com/wp-content/we_transfer/index2.php?email=/", "nb_hyperlinks": 1, "ratio_intHyperlinks": 1, "ratio_extHyperlinks": 0, "ratio_extRedirection": 0, "safe_anchor": 0, "domain_registration_length": 338, "domain_age": 0, "web_traffic":1853, "google_index": 1, "page_rank": 2, }, ]; async function main() { try { // Creating an ONNX inference session with the specified model const model_path = "./models/model.onnx"; const session = await ort.InferenceSession.create(model_path); // Get values from data and remove url links const inputs = urls.map(url => Object.values(url).slice(1)); // Flattening the 2D array to get a 1D array const flattenInputs = inputs.flat(); // Creating an ONNX tensor from the input array const tensor = new ort.Tensor('float32', flattenInputs, [inputs.length, 10]); // Executing the inference session with the input tensor const results = await session.run({"X": tensor}); // Retrieving probability data from the results const probas = results['probabilities'].data; // Displaying results for each URL urls.forEach((url, index) => { // The index * 2 + 1 is used to access the probability associated with the phishing class const proba = probas[index * 2 + 1]; const percent = (proba * 100).toFixed(2); console.log(`URL: ${url.url}`); console.log(`Likelihood of being a phishing site: ${percent}%`); console.log("----"); }); } catch (e) { console.log(`failed to inference ONNX model: ${e}.`); } }; main(); // output: // URL: https://www.rga.com/about/workplace // Likelihood of being a phishing site: 0.89% // ---- ```