KDDM2 Challenge WS 24/25: Time Series Anomaly Detection

This repository contains the submission for the Time Series Anomaly Detection Challenge of Group 10. The final model and the respective source code is located in the directory final_model, all models tested can be found under models.

The models were evaluated against custom test cases, including mostly naive tests as well as some advanced cases. The evaluation and test cases can be located in the directory evaluation

Content of this repository

• /final_model: Contains the final model and the respective source code
• /models: Contains all models tested during the development
• /evaluation: Contains the evaluation of the models
• /data: Contains the input data and the generated test data
• generate_test_files.py: Script to generate test data
• anomaly_detector.py: Class to test models against test data
• requirements.txt: Required packages to run the code

Guide for setting up the environment on Windows

To ensure that all files will run correctly, please use follow this guide.

1. Optional: Create a virtual environment using the following command:

python -m venv <name_of_virtual_environment>

2. Activate the virtual environment using the following command:

<name_of_virtual_environment>\Scripts\activate

3. Install the required packages using the following command:

pip install -r requirements.txt

Guide for creating the executable

The executable in final_model was created by using the following command:

pip install pyinstaller

pyinstaller --onefile final_model.py

This creates an executable in a output directory. To run the executable on Windows, navigate to the directory and use the following command:

start ./final_model <path_to_input_file>

To regenerate the test-files for the evaluation, run the following command:

python generate_test_files.py

This will create the test data based on the normal input data and store the anomal data under data/generated_tests, as well as the respective labels under data/ground-truth.

Testing the models

Each model implemented in the models directory can be tested using the functions provided by the AnomalyDetector class located in anomaly_detector.py. To run a modela against all provided tests, simply run the following command:

python <name_of_model>.py 

Running the evaluation

To run the evaluation, navigate to the evaluation directory use the evaluation.ipynb. A summary of the PATE Scores of models regarding each test case can be found in the results.csv. The evaluation features:

• PATE score calculation of all tested models
• Sensitivity Analysis of KNN Parameters

Authors

• Maria Seiser, seiser@student.tugraz.at
• Vinayak Lal, vinayak.lal@student.tugraz.at
• Bolis Hakim, bolis.hakim@student.tugraz.at