The main goals of Task 4 are the exploration of heterogeneous urban data, and development of algorithms to learn descriptive models for enhancing the current understanding of mobility in Lisbon and support mobility planning decisions. In the context of this task, these goals will be pursued in the absence of situational context data (to be explored in Tasks 6-8).
Two major activities will be pursued:
A1) Comprehensive descriptive analytics
This activity will rely on state-of-the-art principles on descriptive analytics from heterogeneous urban data. Special attention will be given to the spatio-temporal nature of data and problem. In this context, A1 will primarily pursue the following goals:
(a) get intuition from data, including the underlying statistical distributions;
(b) characterize the observed noise (categorizing the major types of noise and their degree);
(c) offer simplistic visual analytics of multi-modal urban data determine the need for additional sensors in order to promote analytics with higher statistical significance;
(d) make data summaries;
(e) test hypothesis for evaluating the statistical significance of specific traffic dynamics;
(f) extract statistical features and explore correlations.
A2) Mining of actionable and non-trivial traffic patterns
This activity aims to discover non-trivial (yet meaningful and relevant) relations from mobility data, including:
(a) spatiotemporal patterns (by extending well-established principles from event data analysis and temporal pattern mining);
(b) deviant behavior (given by unexpected distributions of traffic); and
(c) periodicities and emerging trends.
In addition to pattern-centric views, we will attemptively explore alternative learning methods able to offer a generative view of car traffic data, such as neural networks and dynamic Bayesian networks, to guarantee a more comprehensive and noise-tolerant description of circulation dynamics.
Integration will be consider at the data level (data fusion), learning level and/or post-processing level to guarantee the output of comprehensive patterns of city intermodal mobility.
The expected results of A2 are therefore algorithms for mining frequent, deviant, periodic and emerging traffic patterns of traffic, and learning comprehensive generative models of city mobility.
This task builds upon the data collected and processed in task 2, and provides an essential knowledge-basis for the upcoming Tasks 4 to 7.
INESC-ID will be responsible for the execution of this task, aided by LNEC and the validation efforts of CML and its mobility partners. INESC-ID has extensive expertise on learning descriptive models from complex data structures, including advanced pattern discovery from multivariate time series and multi-attribute event data (Henriques, 2015b; Henriques, 2015a). BM-2 and BM-3 grant holders will also contribute to this task.
Two major activities will be pursued:
A1) Comprehensive descriptive analytics
This activity will rely on state-of-the-art principles on descriptive analytics from heterogeneous urban data. Special attention will be given to the spatio-temporal nature of data and problem. In this context, A1 will primarily pursue the following goals:
(a) get intuition from data, including the underlying statistical distributions;
(b) characterize the observed noise (categorizing the major types of noise and their degree);
(c) offer simplistic visual analytics of multi-modal urban data determine the need for additional sensors in order to promote analytics with higher statistical significance;
(d) make data summaries;
(e) test hypothesis for evaluating the statistical significance of specific traffic dynamics;
(f) extract statistical features and explore correlations.
A2) Mining of actionable and non-trivial traffic patterns
This activity aims to discover non-trivial (yet meaningful and relevant) relations from mobility data, including:
(a) spatiotemporal patterns (by extending well-established principles from event data analysis and temporal pattern mining);
(b) deviant behavior (given by unexpected distributions of traffic); and
(c) periodicities and emerging trends.
In addition to pattern-centric views, we will attemptively explore alternative learning methods able to offer a generative view of car traffic data, such as neural networks and dynamic Bayesian networks, to guarantee a more comprehensive and noise-tolerant description of circulation dynamics.
Integration will be consider at the data level (data fusion), learning level and/or post-processing level to guarantee the output of comprehensive patterns of city intermodal mobility.
The expected results of A2 are therefore algorithms for mining frequent, deviant, periodic and emerging traffic patterns of traffic, and learning comprehensive generative models of city mobility.
This task builds upon the data collected and processed in task 2, and provides an essential knowledge-basis for the upcoming Tasks 4 to 7.
INESC-ID will be responsible for the execution of this task, aided by LNEC and the validation efforts of CML and its mobility partners. INESC-ID has extensive expertise on learning descriptive models from complex data structures, including advanced pattern discovery from multivariate time series and multi-attribute event data (Henriques, 2015b; Henriques, 2015a). BM-2 and BM-3 grant holders will also contribute to this task.