Practical Course Artificial Intelligence: Outbreak Detection

Praktikum aus Künstlicher Intelligenz (6 CP)

Content

An outbreak refers to the rapid spread of an infectious disease, with an epidemic usually referring to a more severe and intense form.
The monitoring and surveillance of (especially contagious) diseases and the identification of current or future outbreaks is of high importance for ensuring the public health. Data mining and machine learning with their different methods can contribute to the support of public health authorities in performing this sensitive task.
In this practical course, we will investigate the usage of different data mining techniques to detect outbreaks and other outbreak related events. We will be working on a real data set provided by the Robert-Koch-Institute which provides information for Germany about infected cases over time and locations. This data set may also be combined with external data sources like weather information.
Outbreak detection involves different tasks. For instance, it includes the nowcasting of outbreaks but also the prediction of distributions of expected disease cases (forecasting). We will also put a special interest on interpretable models which allow to find valuable human-understandable relationships in the data which are relevant for outbreak detection.
Participants will explore and apply different techniques from data mining and machine learning such as time series analysis, outlier detection, auto-regression, weak learning and interpretable machine learning techniques. But also data preprocessing, including data cleansing and homogenization, are necessary tasks when working with real data.

The work will be done in small groups of 4 to 5 students.

Prerequisites and Registration

Completion of a lecture in machine learning or data mining. Practical experience with data mining is helpful, but can also be acquired independently.

Please understand that we can only accept a maximum of 25 students. If there are more interested parties, we unfortunately have to let chance decide. It is indispensable that you register in TUCAN before the day of the kick-off meeting and that you assist to the kick-off meeting.

For further questions feel free to send an email to praktikum@ke.tu-darmstadt.de (The email did not work for some people. Please use eneldo@ke.tu-darmstadt.de in this case) No prior registration is needed besides signing-up in TUCAN. However, please stlll send us an email so that we are able to estimate beforehand the number of participants, and have your E-mail address for possible announcements.

Who, when and where?

The kick-off meeting will be held online on Thursday 23.4. at 16:30. Information about the modalities will be given on time.

The plan for the regular meeting is to take place approximatelly every two or three weeks on Thursdays, 16:30 in A313.

Tentative Schedule

Evaluation

The solution will be created in small groups. Your commitment to the course and the quality of your solution will be assessed. In addition, there will be presentations and written submissions, which will also be included in the evaluation.

Forum

The forum is intended to serve as a platform for participants to ask questions and exchange ideas and results.

Literature

"Improving Outbreak Detection with Stacking of Statistical Surveillance Methods" (M. Kulessa et al., 2019)
"An improved algorithm for outbreak detection in multiple surveillance systems" (A. Noufaily et al., 2012)
"Statistical Challenges Facing Early Outbreak Detection in Biosurveillance" (G. Shmueli et al., 2010)
"Monitoring Count Time Series in R: Aberration Detection in Public Health Surveillance" (Salmon et al., 2016)
"Spatio-Temporal Analysis of Epidemic Phenomena Using the R Package surveillance" (Meyer el al., 2017)

Tools

Some useful tools around outbreak detection and machine learning:

scikit-learn: Framework for machine learning (Python)
surveillance: Implementation of various statistical outbreak detection algorithms (R)
epysurv: Wrapper around the R surveillance package (Python)
EpiSignalDetection: R package for monitoring and visualizing infectious disease surveillance data