Login |
 
 

Overview

Hours per week : 2

Credits : 4

Participants: Min: 2 Max: 12 Expected: 6

Type of course: Seminar  (Bachelor/Master)

Language: english

Content

Let's do the Time Warp - How to Mine Time.

Traditional data mining approaches don't consider time. They are also not able to deal with the information that could be gained from time.  And yet the temporal aspect is very important in many areas.

For example: detecting association rules in shopping baskets. In winter different combinations of products are bought in comparison to summer. Consider twitter as one big stream of information. Can we learn, based on twitter, if an important news came up, or a catastrophe is happening? Consider a manual of a baking process, where many different tasks has to be executed in some distinct order. Can we learn the correct manual just by looking at some labelled(good and bad) instances? Other examples are clusters changing over time, time-based outliers or stock-market analysis. In order to tackle these problems, these approaches need to take multiple new requirements into account.

In this seminar we will be introducing and discussing the different challenges found in time-related data mining.

Temporal data mining can roughly be divided into two parts:
Offline learning, such as analyzing time series, and online learning as it is applied to data streams.

A time series is a finite sample of points or intervals. They are ordered based on time. For data miners, it is interesting just how similar these time series are, or what we can learn from the time series, for example.

In the data stream area, on the other hand, there is one infinite data stream, which can, of course, be of high dimensionality. The interest here is focused on when the stream changes and how we can predict the future behavior of the stream.

Each student is required to give a 30-minute talk about one challenge in temporal data mining and submit a course paper at the end of the semester. The presenter has to discuss the slides for the presentation with her/his advisor (Iris or Sebastian) at least 7 days (hard deadline) before the seminar date.

The oral presentation counts towards 60% of the final mark and the course paper 40%

 

Form of Exam

oral presentation (30 minutes) and written examination at the end of semester

 

Preliminaries

none, but attending Data Mining 1 in parallel is beneficial.