Socio-demographic information is usually only accessible at relatively coarse spatial resolutions. However, its availability at thinner granularities is of substantial interest for several stakeholders, since it enhances the formulation of informed hypotheses on the distribution of population indicators. Spatial disaggregation methods aim to compute these fine-grained estimates, often using regression algorithms that employ ancillary data to re-distribute the aggregated information. However, since disaggregation tasks are ill-posed, and given that examples of disaggregated data at the target geospatial resolution are seldom available, model training is particularly challenging. We propose to address this problem through a self-supervision framework that iteratively refines initial estimates from seminal disaggregation heuristics. Specifically, we propose to co-train two different models, using the results from one model to train/refine the other. By doing so, we are able to explore complementary views from the data. We assessed the use of co-training with a fast regressor based on random forests that takes individual raster cells as input, together with a more expressive model, based on a fully-convolutional neural network, that takes raster patches as input. We also compared co-training against the use of self-training with a single model. In experiments involving the disaggregation of a socio-demographic variable collected for Continental Portugal, the results show that our co-training approach outperforms alternative disaggregation approaches, including methods based on self-training or co-training with two similar fully-convolutional models. Co-training is effective at exploring the characteristics of both regression algorithms, leading to a consistent improvement in different types of error metrics.

With urban development in cities, shared bicycle systems are increasingly used as a way to avoid traffic caused by cars, promoting sustainable mobility and contributing for traffic and pollution reduction in urban areas. The imbalance in the availability of bicycles and docks at the stations of the systems makes it impossible to rent and return bicycles, making it necessary to redistribute them across the network. However, this process has flaws, mainly during rush hours. In this paper, we analyse data provided by the Lisbon City Council regarding their bike sharing system, which has the rebalancing operations' influence. Since the original data was contaminated with the rebalancing operations, an analysis was conducted in an attempt to remove this influence from the data. Following this analysis, a new dataset was created using only the trip data to enable model development for each station and predict the bicycle demand. The plateaus in the created dataset were then analysed to determine if they're due to lack of demand from costumers, or due to stations being full or empty.

Ahstract- The occurrence of seasonal natural phenomena depends on the conditions leading to it and not directly on the progression of time, meaning its context varies across time and space. Examples of this include comparing plant growth, insect development or wildfire risk during the same time period at different locations or in different time periods at the same location. However, visualizing and comparing such phenomena usually implies plotting it across the time axis as it's perceived as temporal data. Since it's not directly dependent of time, identifying patters of recurrence using this technique is inefficient. Because of this, we proposed transforming (when needed) the dependent function to a non-decreasing monotone one, in order to preserve the monotonic property of time progression. Then we used the resulting function as a time axis replacement to achieve an equal ground of comparison between the different contexts in which the phenomenon occurs. We applied this technique to real data from seasonal natural phenomena, such as plant and insect growth, to compare its progression in different temporal and spatial contexts. Since the dependent function of the phenomenon was scientifically known, we were able to directly use the technique to infer its seasonality patterns. Furthermore, we applied the technique to real data from the coronavirus worldwide pandemic by hypothesizing its dependent function and analysing if it was able to reduce the existing temporal misalignment between different contexts, like years and countries. The results achieved were positive, although not as remarkable as when the dependent function was known.

Road traffic inside cities is responsible for noise and pollution, that causes health problems, fuel consumption and waste of time in jams. Mitigation solutions are usually used to soften the impact of this problem in most cities. In particular, the city of Lisbon has taken measures to reduce pollution by closing areas of the city to the most polluting cars - the zero emission zones. However, the city still lacks visual analytics support for traffic decisions in real-time. In this paper we present a traffic flow indicator that can indicate the road traffic fluidity inside a region of interest for a given time frame, and integrated it into a interactive dashboard supported by a predictive model. With this solution, decision makers can analyse historical data and predict short-term traffic behaviour.

Air quality is monitored using data recollected using fixed selected stations in a region, generally a city. Such approach does not support a fine-grained comprehension about the air quality, namely, in areas distant from the collector's stations, specially in residential urban places. In this paper, we describe a platform that will provide to city council decision-makers a visualization of air pollution data, using an interactive map-based dashboard with multiple spatial resolution. The air quality data is collected using low-cost portable sensors. Air pollution data is then integrated with other environmental contextual data and displayed into the dashboard. Such data includes, among other, spatio-temporal mobility data, providing contextual information about air pollution. The solution is tailored to city council decision-makers enabling a better understanding of air quality issues, and acting as a supporting tool for different communities, exploiting synergies to promote the sustainability of the city.

In this paper, the application of different machine learning strategies for current- and vibration-based detection of bearing faults in squirrel-cage induction motors is studied. This study compares several feature extraction strategies such as a statistical and spectral analysis of vibration, a statistical analysis of the Hilbert's Transform envelope of vibration, an analysis of the currents deviation to a perfect sinusoid and a statistical and spectral analysis of the Park's Vector Modulus, with its performances being evaluated with the Support Vector Machine, Artificial Neural Network, Random Forests and Extreme Gradient Boosting algorithms. A comparison of results obtained using sampling frequencies of 0.8 kHz, 1 kHz, 2 kHz, 5 kHz and 10 kHz and analysis periods between 20 ms and 100 ms is made and promising models are achieved even with the lowest sampling frequencies.

Crimes, forest fires, accidents, infectious diseases, human interactions with mobile devices (e.g., tweets) are being logged as spatiotemporal events. For each event, its geographic location, time and related attributes are known with high levels of detail (LoDs). The LoD plays a crucial role when analyzing data, enhancing the user's perception of phenomena. From one LoD to another, some patterns can be easily perceived or different patterns may be detected. Modeling phenomena at different LoDs is needed, as there is no exclusive LoD at which data can be analyzed.Current practices work mainly on a single LoD, driven by the analysts perception, ignoring the fact that the identification of the suitable LoDs is a key issue for pointing relevant patterns.This paper presents a Visual Analytics approach called VAST, that allows users to simultaneously inspect a phenomenon at different LoDs, helping them to see in what LoDs patterns emerge or in what LoDs the perception of the phenomenon is different. In this way, the analysis of vast amounts of spatiotemporal events is assisted, guiding the user in this process.The use of several synthetic and real datasets allowed the evaluation of VAST, which was able to suggest LoDs with different interesting spatiotemporal patterns and the type of expected patterns.

Some phenomena, such as crimes in a city, fires occurred in a country and road accidents can be interpreted as sets of spatio-temporal events. A spatio-temporal event is described by a geographic location, a time instant and other characterising attributes. The cartographic visualisation of spatio-temporal events remains unresolved, due to the challenges related with portraying multiple dimensions simultaneously: the spatial, the temporal and the semantic (zero or more dimensions) phenomenon's components. In this context, this article presents the Attenuation and Accumulation Maps (AA-Maps). The main idea of this visualisation analytic approach consists in showing in a map, the resulting effect of combining attenuation and accumulation, from a temporal reference of observation, given a spatio-temporal Level of Detail (LoD). Imagine the footprints of people crossing a garden in various directions. They leave different traces that summarize the cumulative effect of the footprints on the grass, which is attenuated as time goes by. AA-Maps support different combinations of attenuation and accumulation functions. In addition, this method also enables analysis with different Levels of Detail (LoD), both spatial and temporal. This allows distinct analytic perspectives of the phenomenon while promoting the search for the most suitable parametrization for its characteristics.

The field of Information Visualization seeks to identify the general principles of visualization, and makes use of these principles to propose new forms of visualization for specific types of data. Included in these different data types, it was identified a data type which was not thoroughly explored. The notion of entities where each entity is composed by different attributes, and one of these attributes is composed by one picture which invokes a strong feeling of familiarity to the user. The information that we are attempting to visualize is the most basic type of data structure, a table, where the number of entities to visualize should be higher than we can humanely count, yet smaller than a few thousands. In the field of visualization we identified a niche where the main focus is the image, and despite that it has a vast number of applicable scenarios, it hadn't been properly explored. One of the major attempts at doing so, was by Microsoft Live Labs and it demonstrated limitations that will be addressed by our approach. In order to evaluate the proposed forms of visualization they will be applied and evaluated with the Deloitte Portugal organizational case-study.

Nowadays, organizations are facing several challenges when they try to analyze generated data with the aim of extracting useful information. This analytical capacity needs to be enhanced with tools capable of dealing with big data sets without making the analytical process a difficult task. Clustering is usually used, as this technique does not require any prior knowledge about the data. However, clustering algorithms usually require one or more input parameters that influence the clustering process and the results that can be obtained. This work analyses the relation between the three input parameters of the SNN (Shared Nearest Neighbor) algorithm and proposes specific guidelines for the identification of the appropriate input parameters that optimizes the processing time.

Sharing and integrating information across multiple autonomous and heterogeneous data sources has emerged as a strategic requirement in modern business. We deal with this problem by proposing a declarative approach based on the creation of a reference model and perspective schemata. The former serves as a common semantic meta-model, while the latter defines correspondence between schemata. Furthermore, using the proposed architecture, we developed an inference mechanism which allows the (semi-) automatic derivation of new mappings between schemata from previous ones. The aim of this paper is to present the proposed inference mechanism.

A tuple of fuzzy constraints defined on a discrete satisfaction scale defines a lattice structure, called relaxation space, made of the tuples of level cuts of the fuzzy constraints, equipped with the partial order induced by the scale ordering. Then choosing a way of combining the fuzzy constraints has two joint effects: usually a completion of the ordering between tuples of levels, and possibly a simplification of the relaxation space (by eliminating subsumed tuples corresponding to the same level of global satisfaction). The paper investigates the possibility of specifying the ordering between tuples of level cuts constraints on a simplified relaxation space, without resorting to the use of explicit aggregation operations. The idea is to elicit the preferences between the tuples of constraints representing various relaxations of the set of fuzzy constraints, directly from the user. The way to express these preferences in a local manner is also discussed.