Timo Nolle

nameTimo Nolle
positionResearch Assistant
eMailnolle (AT)tk(DOT)tu-darmstadt(DOT)de
phone+49 (6151) 16 - 23192
fax+49 (6151) 16 - 23202
officeS2|02 B102
postal addressTU Darmstadt - FB 20
FG Telekooperation
Hochschulstraße 10
D-64289 Darmstadt
Germany

Research Interests

  • Deep Learning
  • Machine Learning
  • Artificial Intelligence
  • Data Science

Publications

Displaying results 1 to 3 out of 3

Process Compliance Checking using Taint Flow Analysis
Alexander Seeliger, Timo Nolle, Benedikt Schmidt, Max Mühlhäuser
In: Proceedings of the 37th International Conference on Information Systems (ICIS), vol. 37, p. 1-18, December 2016.
http://aisel.aisnet.org/icis2016/DataScience/Presentations/6/.

Unsupervised Anomaly Detection in Noisy Business Process Event Logs Using Denoising Autoencoders
Timo Nolle, Alexander Seeliger, Max Mühlhäuser
In: Calders, Toon; Ceci, Michelangelo; Malerba, Donato: Discovery Science: 19th International Conference, DS 2016, Bari, Italy, Proceedings, p. 442-456, Calders, Toon Ceci, Michelangelo Malerba, Donato, October 2016. ISBN 978-3-319-46307-0.
http://dx.doi.org/10.1007/978-3-319-46307-0_28.

Data-driven Detection of Congestion-affected Roads
Timo Nolle, Immanuel Schweizer, Frederik Janssen
2014.

[Export this query result to XML]

Theses

1 Entries found


01.12.2016

MyNextDay: A Hybrid Approach for Predicting Complex Student Activities

Master Thesis

finished


Emerging anticipatory applications that intervene or proactively support users in daily life require an in-depth understanding of human behavior, and especially, having knowledge about future states is essential for the development of such proactive systems, termed anticipatory mobile computing. Current state-of-the-art prediction approaches are based on either user routines or next place prediction to accurately predict user's future activities or locations. However, each approach reveals individual drawbacks for comprehensive prediction models. In this thesis, we present a novel approach, namely MyNextDay, to predict human activities by utilizing temporal and spatial features predicted from likely user routines and in between fill up remaining uncertain time slots by sequences of next place prediction. For that, we first specify  user's highly probable routines on a daily and weekly basis utilizing spatiotemporal calendar events as additional source. Next, we utilize a sequence of next place predictions, where the outcome of one prediction is used as input for the next chronological prediction. In the last step, based on these predicted spatial and temporal features, we forecast the activities for an entire day. We capture the data quality of a given four-week user-annotated dataset from 33 participants and, as a result, we obtain user-annotated data for 86.5% of the observed time period. Applying our approach on the dataset, the results show a F1-score  up to 79,8% (7-class problem) for the best performed user using 24-fold cross-validation. We can observe that our approach outperforms baseline classifiers with majority and random vote strategy over all users and achieves consistently better results  than user routine and next place prediction approaches in isolation. We also discuss the cold-start problem for individual models and show that we achieve adequate results after four days. The results open a novel way for a lot of ubiquitous applications such as smart assistants for time management or home automation systems for controlling air conditioning, robotic vacuum cleaner or heating.


A A A | Drucken Print | Impressum Impressum | Sitemap Sitemap | Suche Search | Kontakt Contact | Webseitenanalyse: Mehr Informationen
zum Seitenanfangzum Seitenanfang