AnomalyDetection-Java

Introduction

Java implemented anomaly monitoring algorithm library.

AnomalyDetection Java is an anomaly detection algorithm library based on statistical and machine learning approaches, implemented in Java. It can be easily embedded in actual business to monitor various types of anomalies in data sequences, including outlier anomalies, fluctuation anomalies, trend anomalies, and so on.

Feature

Its main feature are as follows:

Depends on JDK environment and common-math3 library, and conducts graphical testing samples through JFreeChart.

High scalability: The architecture is very simple. Developers can flexibly and efficiently add business domain anomaly detection rules, or add other types of algorithms. The following monitoring algorithms are currently supported:

Exception type	Algorithm	Principle	Limitation
Absolute value anomaly	GESD	Calculate generalized extreme student bias statistic to find outliers	Apply normal distribution
Absolute value anomaly	Z-score	Z-Score based detection	Apply normal distribution
Absolute value anomaly	Quantile	Quantile based statistical algorithm	Applicable to all distributions
Abnormal fluctuations	2nd derivative+distance	Finding MBP based on second derivative and distance	Normal and non normal distributions of inflection points
Trend anomaly	Mann Kendall	Based on Mann Kendall test	Applicable to all distributions
Threshold anomaly	Threshold rule engine	Based on business threshold rules	Applicable to all distributions

Demos

ad4j-Demos

For Developers

Building ad4j

you can build ad4j using Maven by issuing the following command from the root directory of the project:

mvn clean install -Dmaven.test.skip=true

The build requires JDK 8 or later.

Using ad4j

add to maven pom:


<dependency>
<groupId>org.algorithmtools</groupId>
<artifactId>ad4j</artifactId>
<version>${version}</version>
<scope>system</scope>
<systemPath>${project.basedir}/lib/ad4j-${version}.jar</systemPath>
</dependency>

- business use:
  `BizUseExample`
```java
public class BizUseExample {

    public static void main(String[] args) {
        indicatorDetect();
    }

    public static void indicatorDetect(){
        // 1. Transfer biz data to indicator series info
        long currentTime = System.currentTimeMillis();
        List<IndicatorSeries> indicatorSeries = new ArrayList<>();
        indicatorSeries.add(new IndicatorSeries(currentTime + 1, 1d, "logicalIndex-1"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 2, 2d, "logicalIndex-2"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 3, 3d, "logicalIndex-3"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 4, 4d, "logicalIndex-4"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 5, 40d, "logicalIndex-5"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 6, 6d, "logicalIndex-6"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 7, 7d, "logicalIndex-7"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 8, 8d, "logicalIndex-8"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 9, 9d, "logicalIndex-9"));
        indicatorSeries.add(new IndicatorSeries(currentTime + 10, 10d, "logicalIndex-10"));

        IndicatorInfo info = new IndicatorInfo("Example", "Example-Name", indicatorSeries);

        // 2. New AnomalyDetectionEngine to detect
        AnomalyDetectionEngine engine = new AnomalyDetectionEngine();
        AnomalyDetectionResult detectionResult = engine.detect(info);

        // 3. Business process detect result. Like Records,Alarms,Print
        IndicatorSeriesUtil.print(detectionResult);
    }

}

Test case

ADEngineTest.java

Indicator original data:10.0, 12.0, 12.5, 133.0, 13.0, 10.5, 100.0, 14.0, 15.0, 14.5, 15.5
This is Line chart:

ADEngineTestprint result:

==============Anomaly Detection Result=============
1.Anomaly detection original data:[[0, 10.0, 0], [1, 12.0, 1], [2, 12.5, 2], [3, 133.0, 3], [4, 13.0, 4], [5, 10.5, 5], [6, 100.0, 6], [7, 14.0, 7], [8, 15.0, 8], [9, 14.5, 9], [10, 15.5, 10]]
2.Overview: has 3 types anomaly detected.
3.1.[Anomaly: Fluctuation Anomaly] [{"anomalyDetectionModel":"MODEL_ADM_2ndDerivationMBP","anomalyType":"TYPE_FLUCTUATION","hasAnomaly":true,"normalRangeMax":19.75,"normalRangeMin":7.75,"seriesList":[{"logicalIndex":"3","time":3,"value":133.0}]}]
3.2.[Anomaly: Outliers Anomaly] [{"anomalyDetectionModel":"MODEL_ADM_GESD","anomalyType":"TYPE_OUTLIERS_VALUE","hasAnomaly":true,"normalRangeMax":0.0,"normalRangeMin":0.0,"seriesList":[{"logicalIndex":"3","time":3,"value":133.0},{"logicalIndex":"6","time":6,"value":100.0}]},{"anomalyDetectionModel":"MODEL_ADM_Quantile","anomalyType":"TYPE_OUTLIERS_VALUE","hasAnomaly":true,"normalRangeMax":19.75,"normalRangeMin":7.75,"seriesList":[{"$ref":"$[0].seriesList[0]"},{"$ref":"$[0].seriesList[1]"}]}]
3.3.[Anomaly: Trend Anomaly] [{"anomalyDetectionModel":"MODEL_ADM_ManKendall","anomalyTrend":"TREND_UP","anomalyType":"TYPE_TREND","hasAnomaly":true,"normalRangeMax":0.0,"normalRangeMin":0.0}]
==============Anomaly Detection Result End=========
click Enter to close window...

More monitoring algorithms are being added, welcome to co build.

Participate in Contributions

Welcome to join the community, build a win-win situation, please refer to the contribution process： How to contribute.

Thank you to all the people who already contributed to AnomalyDetection-Java!

Get Help

Create an issue and describe it clearly.

algorithm-tools / ad4j

readme