In this second release in the new “Studies in Big Data” series, Springer provides a timely demonstration of how the “nexus of forces”--the convergence of social interaction, mobility, cloud, and information patterns that drive new business scenarios [1]--can be put to work on small mobile devices such as tablets and smartphones to tackle big data challenges. In their compact yet well-focused monograph, authors Gaber, Stahl, and Gomes provide an introduction to data mining and knowledge discovery using mobile devices, from early work where the user terminal was used as a front-end, to high-performance computing platforms, to more recent developments in which computations are performed on the actual mobile devices.
The pocket data mining (PDM) architecture they propose comprises three different agent types: mobile agent miners (AMs), mobile agent resource discoverers (MRDs), and mobile agent decision makers (MADMs). AMs are charged with performing stream mining and learning functions, both on the local user terminal and on external user terminals, in a distributed way. MADMs can roam through the mobile network to task individual AMs and to collect information or partial results. MRDs are used to perform the discovery of the different agents available on the network (data sources, sensors, AMs and their computational capabilities) and to derive schedules for MADMs. A procedural description of the PDM workflow is provided, together with simulated results from two case studies, using two different learning algorithms implemented in AMs (C4.5 and naïve Bayes).
In addition to discussing simulated results, the authors provide details of their initial PDM implementation using Java on the Android platform, which includes AMs and MADMs, but not MRDs. The absence of the resource discovery service means that the current implementation is neither particularly scalable nor reliable: the PDM network has to be initialized manually, using fixed naming conventions, and there is no automatic recovery in case of failure of one or more AMs.
Collaborative-stream PDM is also introduced, at the conceptual level, to provide an approach in which different models are weighted and selected, using voting methods or algorithmic decisions, based on the local accuracy in addressing the target problem. Promising experimental results are also presented, building on a Java implementation of the learning algorithm, but practical applicability is still far away, as the communication protocols between the different mobile agents, as well as their resource implications, have yet to be defined and still represent an open challenge.
The book concludes with a forward-looking section outlining some possible applications of pocket data mining for decision support in healthcare, financial trading, public safety, and defense.
Pioneering work is described in this book, and a lot of additional research and implementation efforts will be required to see it ready for prime time. Some additional factors that need to be addressed are resource awareness on battery-powered platforms, the effective visualization of results on small displays, and the efficiency and security of communications protocols. This interesting book provides an introduction to this new and promising field. There are some typographical errors here and there, but they do not impede understanding.