Logistic regression

Case 5. Loan approval

In lending tasks, a model rarely answers "yes or no" directly. Instead, it estimates a probability that is then turned into a decision.
Even more important is the client’s risk. This is why logistic regression is used so often in credit scoring: it gives not just a label, but the probability that the client is reliable – something you can work with.

Case Goal
Estimate the client’s credit risk and make a loan approval decision.

The model should:
1) Compute the probability that the client will repay the loan
2) Allow flexible decision control via a threshold

Example of use


                
<?php

use Rubix\ML\Classifiers\LogisticRegression;
use Rubix\ML\Datasets\Labeled;
use Rubix\ML\Datasets\Unlabeled;

$samples = [
    [3000, 600, 0.4],
    [8000, 750, 0.2],
    [2000, 500, 0.7],
    [10000, 800, 0.1],
];

$labels = ['decline', 'approve', 'decline', 'approve'];

$dataset = new Labeled($samples, $labels);

$model = new LogisticRegression();
$model->train($dataset);

$client = new Unlabeled([[5000, 680, 0.3]]);
$prediction = $model->predict($client);

echo 'Predicted label: ' . PHP_EOL;
print_r($prediction);

$probas = $model->proba($client);
$probabilityOfApproval = $probas[0]['approve'] ?? null;

echo  PHP_EOL . 'Probability of approval (class=approve): ';
print_r($probabilityOfApproval);

echo PHP_EOL;

$threshold = 0.6;
$approved = $probabilityOfApproval !== null && $probabilityOfApproval >= $threshold;

echo 'Threshold: ' . $threshold . PHP_EOL;
echo 'Decision: ' . ($approved ? 'APPROVE' : 'DECLINE') . PHP_EOL;

Result: Memory: 1.191 Mb Time running: 0.006 sec.


                Predicted label: 
Array
(
    [0] => decline
)

Probability of approval (class=approve): 0
Threshold: 0.6
Decision: DECLINE