LIAAD/INESC TEC researchers win award in Germany
Luís Matias and Petr Kosina, researchers at INESC TEC’s Laboratory of Artificial Intelligence and Decision Support (LIAAD), won the first prize in the Data Mining competition at the Resource-aware Machine Learning - International Summer School 2012, which took place on 7 September at the Technische Universität Dortmund, Germany. This is a prestigious event known for the quality of the speakers who attend it every year.
A total of 50 students competed in the challenge which was to develop a model that would make it possible to sort data from various smart phones in relation to their next location (base station) with maximum accuracy, and with a minimum number of attributes. The ultimate goal was to build a Mobile Phone Tracking tool, which basically consists of trying to find the next position of the phone/individual (that is, by identifying the nearest base station), based on the data history of each phone.
The idea was trying to find a methodology that would obtain the best accuracy in guessing the next position of the mobile phone with the lowest number of errors possible. This methodology could be composed of preprocessing methods, variable selection and the selection of algorithms to be used for automatic learning and classification. The final score also took into account the number of variables since monitoring the 120 variables of a smartphone is not sustainable and can quickly consume the battery of the phone. Therefore, minimal use of resources was benefitted.
The methodology proposed by the LIAAD researchers used only one set two/three decision variables and an ensemble technique – Stacking – using Bayesian Networks and Decision Trees based on classification rules. With this solution, the researchers obtained the best score in the competition.
Luís Matias is currently doing a PhD in Computer Science at the Faculty of Engineering of the University of Porto (FEUP), and specialises in Machine Learning algorithms for solving problems of Supervised and Unsupervised Classification, Regression and time series analysis in contexts of continuous streams of data. He is also Visiting Lecturer at the DEI-FEUP since September 2009.
Petr Kosina is from the Czech Republic and is currently doing a PhD in Computer Science at the Masaryk University (MU), while developing his research at LIAAD. His research interests include data mining from data streams, namely classification and change detection.
