The importance of meaningful segmentation of consumers from a retail perspective cannot be overstated. Current approaches, however, oversimplifies the situation, most notably by aggregating away time, conflating behaviour based on temporal differences – despite the fact that time, in for example the form of seasons and holidays, has a well known influence on retail behaviour. Addressing this a novel technique has been developed. The approach has since been evaluated and applied internationally within two retail companies, considering both consumer and store segmentation and description. [MORE INFORMATION]
Dar es Salaam suffers from biannual flooding caused by various factors including climate change and rapid unplanned urban growth. Receiving a ESRC-NERC-DFID grant for “Big Data for Flood Resilience in East Africa”, this project investigates how digital footprint data can predict, identify and inform decisions during flood and other natural hazard events. The research informed DFID’s Big Data for Climate Change and Disaster Resilience Report. [MORE INFORMATION]
It has been recently claimed that human movement is ‘highly predictable’ upper bound of 93% predictability shown. However, knowing an upper bound is only useful if it is relatively tight, i.e. it is close to the true limit of predictability. In this work we reconsider the derivation of the upper bound to movement predictability and by considering real-world topological constraints we are able to achieve a tighter upper bound. Our results show that the upper bound is actually between 11-24% less than previously claimed. [MORE INFORMATION]
Time series symbolisation is a common pre-processing step to speed up computation, reduce storage costs and/or enable the application of certain algorithms. In this work we show that current symbolisation techniques are sub-optimal in (at least) the broad application area of time series comparison leading to unnecessary data corruption and potential performance loss before any real data mining takes place. Addressing this we present two novel algorithms which are shown to be optimal under some broadly applicable assumptions. [MORE INFORMATION]
How could donated data help our society? Data is a rich source of people’s habits – shopping data from loyalty cards, for example, can reflect our diet. If people donate their personal data for research, analysis of it can provide scope to improve everything from understandings of the dietary pre-cursors to diabetes to the impact of lifestyle on heart disease. But there are vital issues around the collection and use of personal data that must be addressed. This research project examines a number of these issues discussing the implications of these results in terms of data sharing practices. [Article in The Conversation] [Research Article]
Data streams whose events occur at random arrival times rather than at the regular, tick-tock intervals of traditional time series are highly prevalent, particularly in human behavioural data. Continuous, irregular and often highly sparse, these event series differ greatly to traditional, regularly sampled, time series and traditional approaches to analysis tend to perform poorly. This focus of this project is the development of novel techniques addressing this. Two examples are the extraction of underlying time-frequency patterns in highly intermittent data (Research Article) and the prediction of non-stationary, intermittent time series (submitted).
Maps are crucial in emerging countries to make decisions, but often do not exist. Community mapping is a participatory process where community members map their neighbourhoods, making maps openly available in the process. In emerging nations it is becoming a very useful method for illuminating urban areas of rapid growth and change, providing policy and decision makers maps and data with which to make informed decisions. From a community perspective, it facilitates neighbourhood improvements and action, in some of the most deprived areas of the world. The focus of this project has been Ramani Huria, an effort to map over 1.3 million residents in Dar es Salaam, Tanzania – Africa’s fastest growing city. [MORE INFORMATION]
Land use classification is a critical information to monitor a territory, support its development and plan its evolution. Aerial imagery is generally used as the base media to derive a land use classification by manual digitisation. However, this task has proved to be very time consuming given the generally large extent of the area to be classified. We investigated how state-of-the-art deep learning methods can be used to automate land use classification from aerial imagery. The work focused on the city of Dar-Es-Salaam where both aerial imagery and a land use classification ground truth were available. [Research Article]
Leveraging recent deep learning advances with high resolution drone imagery, the team has developed state-of-the-art techniques for the automatic detection of buildings and roads from aerial imagery. N/LAB is now focusing on the extension of these techniques to also extract road conditions. Such information is typically obtained through on the ground surveying requiring significant human resources. Directly aiming to reduce the human and monetary cost within developing countries, the project aims to directly address the challenges these countries face jointly due to rapid road network evolution and limited resources.
The Projection Augmented Relief Model (PARM) system is an award winning new display system that provides a physical, 3-Dimensional approach to data visualization. Using digital projection onto physical models in combination with a dynamic display ecology the PARM combines the affordances of interactive mapping and physical landscape modelling to promote more effective engagement. [Research Article]