Universiteit Leiden

nl en

Lecture

This Week’s Discoveries | 6 December 2016

Date
6 December 2016
Time
Explanation
This Week's Discoveries
Address
Oort Building
Niels Bohrweg 2
2333 CA Leiden
Room
De Sitterzaal

Lecture

Title
Mining Multivariate Time Series: Methods, Tools and Applications

Speaker
Ricardo Cachucho (LIACS) is a PhD student in the Data mining group at LIACS.  He holds a Master degree in data analytics from the Economics Faculty of Porto. Originally inspired to study time series after studying evolutionary economics, for his PhD he now focusses on data mining techniques for time series.

Abstract
Method 
Fitting sensors to humans and physical structures is becoming more and more common. These developments provide us large multivariate datasets with opportunities for new discoveries using data mining methods. We will present Accordion, an embedded method that constructs and selectsaggregate features iteratively, in a memory-conscious fashion. Our algorithms work on both classification and regression problems. 
Tool 
A main challenge for a method such as Accordion is to reach its users. In most cases, these users are domain experts from study fields without a computer science background. To bridge the gap between users and the method we created a tool that allows users to preprocess, visualise, and learn regression and classification models from multivariate time series. 
Application
The Leiden Longevity Study is designed to identify genetic and phenotype markers that relate to longevity. In order to study the behaviour of the study participants, free living sensor data was collected. A validation study has been collecting data with an annotated set of physical activities labels and multiple sensor systems. As research partner, LIACS is bridging the gap between mining methods and their needs for different  models.