Case 4. MLP for nonlinear phishing risk (RubixML)

Implementation in RubixML

We solve this task with a multilayer neural network. In this case, phishing risk is estimated from two features: employee security-awareness level and current workload. The dependency can be nonlinear, so we use an MLP (Multilayer Perceptron) from RubixML: build a labeled dataset, standardize features with ZScaleStandardizer, train the model, and get a prediction for a new employee profile.

Example of code:


<?php

use Rubix\ML\Datasets\Labeled;
use Rubix\ML\Classifiers\MultilayerPerceptron;
use Rubix\ML\NeuralNet\ActivationFunctions\ReLU;
use Rubix\ML\NeuralNet\Layers\Activation;
use Rubix\ML\NeuralNet\Layers\Dense;
use Rubix\ML\Transformers\ZScaleStandardizer;

$samples = [
    [1, 6],
    [5, 3],
    [10, 1],
    [3, 7],
];

$labels = [
    'risk',
    'safe',
    'risk',
    'safe',
];

$dataset = new Labeled($samples, $labels);
$standardizer = new ZScaleStandardizer();
$dataset->apply($standardizer);

// Hidden layers: 8 and 4 neurons add enough nonlinearity for this tiny demo dataset.
$model = new MultilayerPerceptron([
    new Dense(8),
    new Activation(new ReLU()),
    new Dense(4),
    new Activation(new ReLU()),
]);

$model->train($dataset);