4 Presentations

• 
  10:30 AM - 10:55 AM

  Traffic detection and prediction with CNNs and regression analysis

  • Mohsen Amoei, presenter, Master's Student
  • Anjali Awasthi, Professor

  An essential act for developing smart cities is having a profound understanding of the traffic flow. These objectives could be efficiently realized through Machine Learning applications. Traffic congestion can be viewed as a product of the interaction between demand and capacity. Periodic high-demand at specific bottlenecks during peak hours can result in chronic congestion. Big Data generated from different sources could be leveraged to develop congestion measurement in real time. Our goal is to develop a model to apply computer vision, and data science approaches to detect traffic elements and predict future congestions.
  Keywords: Computer Vision, Traffic Prediction, Data Mining, CNN, Regression Analysis
• 
  10:55 AM - 11:20 AM

  Forecasting demand of container shipments

  • Greta Laage, presenter, Polytechnique Montréal
  • Emma Frejinger, DIRO and CIRRELT
  • Gilles Savard, Polytechnique Montréal

  Freight carriers require accurate demand forecasts to adequately plan their operations. We focus on forecasting short-term demand for container transportation by a rail carrier. We present prediction results of machine learning algorithms trained on real data and we assess the impact of demand forecast accuracy on so-called block plan solutions.
• 
  11:20 AM - 11:45 AM

  Predicting buses end-trip delay using machine learning algorithms to model planning effectiveness

  • Victor Hannothiaux, presenter, Polytechnique Montréal
  • Andrea Lodi, Polytechnique Montreal
  • Louis-Martin Rousseau, Polytechnique Montréal

  Abstract: Service reliability is one of the key quality factors in public transport and resides on the difference between the expected service in time and comfort relative to the one perceived by the user. Therefore modeling bus end-trip arrival time using planning features could be used to assess planning effectiveness and robustness. The proposed method consisted in building and optimizing different machine-learning algorithms (Random Forest, Artificial Neural Network and Gradient Boosted Tree) for multi-label classification of end-trip bus delay. The tests were made on the Montréal public transport network with the support of the Giro company, and the results were therefore compared to classic modeling using probabilistic methods.

  Keywords: Public transport, delay prediction, service reliability, Machine learning
• 
  11:45 AM - 12:10 PM

  Choice modeling for halo effects

  • Claudio Sole, presenter, Polytechnique Montreal
  • Andrea Lodi, Polytechnique Montreal
  • Sanjay Dominik Jena, Université du Québec à Montréal

  Random Utility Maximization (RUM) is arguably the most adopted framework for modeling human choice. This framework, however, is unable to capture complex choice behaviors such as halo effects. This talk reviews the literature on choice models proposed to capture such effects and presents some preliminary results regarding their estimation.

  Choice models, halo effects, Machine Learning