MOTSAD - Multi-Objective Time Series Anomaly Detection

MOTSAD (Multi-Objective Time-Series Anomaly Detection) is a framework that integrates multi-objective optimization with time-series anomaly detection models.

Table of Contents

• Installation
• Datasets
• Quick Start
• Project Structure
• Usage
• Evaluation

Installation

Git clone and enter the folder

cd MOTSAD

Create and activate conda environment

conda create -n motsad python=3.10.16
conda activate motsad

Install dependencies

pip install -r requirements.txt

Datasets

Supported Datasets

1. Server Machine Dataset (SMD)

   • Download from OmniAnomaly
   • Requires preprocessing: bash python preprocess.py

2. AIOPS 2018 Dataset

   • Download from AIOPS

Quick Start

Run MOTSAD algorithm on SMD dataset:

cd lstmad
python motsad.py --dataname machine-1-1

Usage

Running MOTSAD

cd lstmad
python motsad.py --dataname <dataset_name>

Evaluation Metrics

1. Hypervolume Metric:

python evaluation.py --folder lstmad/result/motsad/machine-1-1

2. Pareto Front Computation:

python compute_pareto_front.py --folder lstmad/result/motsad/machine-1-1

Project Structure

MOTSAD/
├── data/
│   └── SMD/
│       └── machine-1-1/
│           ├── machine-1-1_test.csv
│           └── machine-1-1_train.csv
├── lstmad/
│   ├── dataset.py
│   ├── early_stopping.py
│   ├── lstm.py
│   ├── main.py
│   ├── motsad.py
├── compute_pareto_front.py
├── evaluation.py
├── preprocess.py
├── utils.py
├── requirements.txt
├── LICENSE
└── README.md