Vitor Cerqueira

I am a Ph.D. student at the University of Porto, and a research fellow in LIAAD, a laboratory for artificial intelligence and decision support systems. My main research interests are meta-learning and forecasting. My supervisors are Luis Torgo and Carlos Soares.

Email: vitor.cerqueira at

CV  /  Google Scholar  / 

  • 07/2019: Started working on activity monitoring (early detection of rare events in time series). My first paper on this topic was accepted in the International Conference in Discovery Science. The paper is entitled "Layered learning for early anomaly detection: Predicting critical health episodes", and is authored by me and my supervisors.
  • 09/2018: "Arbitrage of forecasting experts" has been accepted in Machine Learning journal, an extension of the work we published in ECML'17.
  • 06/2018: I got a paper accepted in ECML-PKDD 2018: Cerqueira, V., Pinto, F., Torgo, L., Soares, C, and Moniz, N.: Constructive Aggregation and it Application to Forecasting (September, 2018).
  • 02/2018: I did a talk at Farfetch, an online fashion retail platform. The topic was about ensemble methods for time series forecasting, with particular emphasis on the arbitrage approach I've been developing. I hope the data science team at Farfetch enjoyed the presentation!
  • 01/2018: I am now an invited reviewer for T-ITS journal, IEEE Transactions on Intelligent Transportation Systems.
  • 10/2017: I presented my research on forecast combination at Feedzai, one of the most successful companies in fraud detection problems. Thank you for having me and helping me on my research!
  • 09/2017: I published tsensembler on CRAN: an R package for forecast combination. For now, the main method is ADE, Arbitrated Dynamic Ensemble, which is a mixture of experts approach for combining several forecasting models.
  • 08/2017: I was awarded as "Fora de Serie" (outstanding) researcher, a distinguishion by INESC TEC.
  • 08/2017: I got a paper accepted in KNOWMe: 1st International Workshop on Knowledge Discovery from Mobility and Transportation Systems on automated evaluation of Floating Car Data.
  • 07/2017: I got two papers accepted in DSAA 2017 - The 4th IEEE International Conference on Data Science and Advanced Analytics:
    • Dynamic and Heterogeneous Ensembles for Time Series Forecasting
    • A Comparative Study of Performance Estimation Methods for Time Series Forecasting
  • 06/2017: I got a paper accepted in ECML-PKDD 2017, which was awarded the Best Student Machine Learning Paper Award: Cerqueira, V., Torgo, L., Pinto, F., and Soares, C: Arbitrated Ensemble for Time Series Forecasting (September, 2017). In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (pp. 478-494). Springer, Cham. Great job, guys!.
  • 03/2017: I got a paper accepted in the 14th International Work-Conference on Artificial Neural Networks with the paper: Cerqueira, Vitor, L. Torgo, and C. Soares. "Arbitrated ensemble for solar radiation forecasting." International Work-Conference on Artificial Neural Networks. Springer, 2017.
  • 01/2017: I got the Best Paper Award in the Doctoral Sysposium in Informatics Engineering with the paper: V. Cerqueira, and L. Torgo: Combining Forecasters using Arbitration for Water Consumption Forecasting
  • 09/2016: I started my Ph.D. doctoral program in the Faculty of Engineering of the University of Porto
  • 01/2016: I started working with L. Torgo. Our starting topic of research is dynamic ensembles for forecasting.
  • 11/2015: Leaving Germany after an internship in NEC Laboratories. There I worked on Intelligent Transportation Systems with Luis Moreira-Matias and Jihed Khiary.

This area contains selected publications about my research. For a complete publication record check out my Google scholar page.

Arbitrated Ensemble for Time Series Forecasting
V. Cerqueira, L. Torgo, F. Pinto, and C. Soares
Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 2017
Best Student Machine Learning Paper Award

A dynamic ensemble method from the family of mixture of experts geared towards forecasting problems. The learning process of each expert is framed as a predictive task, and the mixing proporting are computed according to loss predictions.

Dynamics of Communities in Large Scale Social Networks
V. Cerqueira, M. Oliveira, and J. Gama
17th International Conference on Enterprise Information Systems (2015)

In my master thesis I studied the dynamics of communities in large scale social networks. The social network was built using CDR data from a major european telecom company (who prefers to remain anonymous :-)). Our methodology encompassed network sampling using a Markov Chain Monte Carlo algorithm (python code available in my github); community detection using Louvain's method; community profiling using RFM analysis, and the analysis of the dynamics of the communities using MECnet.


tsensembler R package for forecast combination
V. Cerqueira
Early version (0.0.5) available on CRAN

In the interest of reproducible research, I am publishing my research in a software package called tsensembler.

The purpose of the package is forecast combination. Particularly, the main method available is ADE, described above. I will be adding more stuff as my research unfolds.

  • Invited reviewer for AAAI Conference on Artificial Intelligence
  • Invited reviewer for TKDE, IEEE Transactions on Knowledge and Data Engineering
  • Invited reviewer for T-ITS, IEEE Transactions on Intelligent Transportation Systems
  • Invited reviewer for Journal of Ambient Intelligence and Humanized Computing (AIHC)
  • Reviewer for the International Workshop on Learning with Imbalanced Domains: Theory and Applications