Anomaly Detection in Time Series 4010-ATSa-30
A detailed description of the subject is provided at the level of a given teaching cycle in the Description section.
|
Term 2026Z:
The lecture will be divided into three parts: (i) time series analysis, (ii) anomaly detection, and (iii) anomaly detection in time series. Each part will cover important ideas and practical skills. In the first part, students will learn what a time series is and why it matters in fields like finance, healthcare, and engineering. They will understand how to break down time series data into trends, seasonal patterns, and residuals. Students will also explore different models for time series forecasting, including additive, multiplicative, and pseudo-additive models, and how to apply these models using Python. Students will learn how to tell the difference between stationary and nonstationary time series. Data smoothing techniques, from simple moving averages to more complex triple exponential smoothing, will be covered. The importance of autocorrelation and partial autocorrelation functions will be discussed, as well as how to use them in modeling. Additionally, students will be introduced to advanced methods like recurrent neural networks (RNNs) and long short-term memory (LSTM) networks for time series forecasting. In the second part, the focus will be on anomaly detection. Students will learn about different types of anomalies, such as point, contextual, and collective anomalies, and their applications in areas like fraud detection and network security. Algorithms for detecting anomalies, such as support vector machines (SVMs), local outlier factor (LOF), k-nearest neighbors (KNN), and k-means clustering, will be explored. The course will also address the challenges of working with high-dimensional data, including techniques like subspace methods, feature bagging, and isolation forests. The final part will combine the knowledge from the first two sections. Students will apply time series analysis techniques to anomaly detection problems. Practical skills will be developed through lab sessions where students work on projects involving time series data and anomaly detection. All examples and techniques will be implemented in Python, giving students hands-on experience with real-world data. |
Course coordinators
Type of course
Mode
Prerequisites (description)
Assessment criteria
1. Completion of the course ends with one common grade for all classes (lecture + laboratory) carried out as part of the course.
2. A sick leave does not exempt students from knowledge of the material. It only entitles them to an individualized form of assessment.
3. Students who have received approval for an individualized course of study are required to contact the course coordinator to determine how to achieve all the learning outcomes assigned to the course. If the above-mentioned outcomes cannot be achieved, the coordinator may refuse to grant credit for the course.
4. Attendance is mandatory. In cases of justified absences, the student is required to contact the course coordinator immediately.
5. Regarding the use of artificial intelligence (AI) tools in the preparation of coursework, the rules set out in the "Regulations on the Use of Artificial Intelligence Tools in the Educational Process for the Computational Engineering Degree Programme" apply.
6. Detailed grading criteria are provided at the level of the given academic cycle in the "Notes" section.
Practical placement
Not applicable.
Bibliography
The subject literature has been posted at the level of a given teaching cycle in the Literature section.
|
Term 2026Z:
materials and courses available online. |
Notes
|
Term 2026Z:
Assesment methods and criteria: Students receive credit for classes based on: The final grade for the course is based on the grade from the final exam. |