Anomaly Detection and Outlier Handling Using PHP and Machine Learning

Overview: In practical data processing, outliers often appear due to measurement errors, unpredictable events, or data source issues, negatively affecting analysis, model training, and prediction. This article introduces how to use PHP and machine learning techniques for anomaly detection and outlier handling.

1. Anomaly Detection Methods

Various machine‑learning algorithms can be used to detect outliers. Two common methods are presented:

1.1 Z-Score Method

The Z-Score method is a statistical approach that calculates each data point’s deviation from the dataset mean. Steps: compute mean and standard deviation; calculate deviation = (data‑mean)/std for each point; flag points with deviation greater than a threshold (commonly 3) as outliers.

function zscore($data, $threshold){
    $mean = array_sum($data) / count($data);
    $std = sqrt(array_sum(array_map(function($x) use ($mean) { return pow($x - $mean, 2); }, $data)) / count($data));
    $result = [];
    foreach ($data as $value) {
        $deviation = ($value - $mean) / $std;
        if (abs($deviation) > $threshold) {
            $result[] = $value;
        }
    }
    return $result;
}

$data = [1, 2, 3, 4, 5, 100];
$threshold = 3;
$result = zscore($data, $threshold);

echo "异常值检测结果：" . implode(", ", $result);

1.2 Isolation Forest

Isolation Forest builds random binary trees to isolate data points; shorter path lengths indicate anomalies. Steps: randomly select a feature and split point; recursively partition data until each leaf contains a single point or max depth is reached; compute anomaly score from path length.

require_once('anomaly_detection.php');

$data = [1, 2, 3, 4, 5, 100];
$contamination = 0.1;
$forest = new IsolationForest($contamination);
$forest->fit($data);
$result = $forest->predict($data);

echo "异常值检测结果：" . implode(", ", $result);

2. Outlier Handling Methods

After detection, outliers can be processed. Two common approaches are shown:

2.1 Remove Outliers

A simple method is to delete outliers based on detection results. Sample code demonstrates filtering values whose absolute magnitude exceeds a threshold.

function removeOutliers($data, $threshold){
    $result = [];
    foreach ($data as $value) {
        if (abs($value) <= $threshold) {
            $result[] = $value;
        }
    }
    return $result;
}

$data = [1, 2, 3, 4, 5, 100];
$threshold = 3;
$result = removeOutliers($data, $threshold);

echo "异常值处理结果：" . implode(", ", $result);

2.2 Replace Outliers

Alternatively, replace outliers with a reasonable value such as the mean or median to preserve dataset distribution. Sample code shows substituting values that exceed a threshold with a replacement value.

function replaceOutliers($data, $threshold, $replacement){
    $result = [];
    foreach ($data as $value) {
        if (abs($value) > $threshold) {
            $result[] = $replacement;
        } else {
            $result[] = $value;
        }
    }
    return $result;
}

$data = [1, 2, 3, 4, 5, 100];
$threshold = 3;
$replacement = 0;
$result = replaceOutliers($data, $threshold, $replacement);

echo "异常值处理结果：" . implode(", ", $result);

Conclusion

The article presented PHP‑based implementations of Z‑Score and Isolation Forest for anomaly detection, and demonstrated removal or replacement techniques for handling detected outliers, helping to clean data, improve model accuracy, and enable more reliable analysis and prediction.