suedaglgn / novelty-detector-app

Novelty Detection on Time Series Data using Gaussion Distribution and CatBoost

Novelty Detection App on Time Series Data

Data

ambient_temperature_system_failure.csv from The Numenta Anomaly Benchmark (NAB)

Techniques

• EllipticEnvelope
• CatBoost

Usage

1- Run ipynb file to create model

2- Build docker file in project directory

docker build -t novelty_detector .

3- Run docker image

docker run -p 5000:5000 novelty_detector

4- Use postman or curl to predict

Curl:

curl --location --request POST 'http://127.0.0.0:5000/predict' \
--header 'Content-Type: application/json' \
--data-raw '{
    "value": 13
}'

Json:

{
    "value": 13
}

API docs

Endpoints	Method	Params	Definition	Sample Value	Sample Response
/predict	POST	value	Value that want to predict	13	{'prediction': [1]}
/report	GET	-	-	-	Number of normal values and Number of anormal values

Result

              precision    recall  f1-score   support

           0       1.00      1.00      1.00      1797
           1       1.00      0.90      0.95        20

    accuracy                           1.00      1817
   macro avg       1.00      0.95      0.97      1817
weighted avg       1.00      1.00      1.00      1817