Balancing energy, latency and accuracy for mobile sensor data classification

Abstract

Sensor convergence on the mobile phone is spawning a broad base of new and interesting mobile applications. As applications grow in sophistication, raw sensor readings often require classification into more useful application-specific high-level data. For example, GPS readings can be classified as running, walking or biking. Unfortunately, traditional classifiers are not built for the challenges of mobile systems: energy, latency, and the dynamics of mobile. Kobe is a tool that aids mobile classifier development. With the help of a SQL-like programming interface, Kobe performs profiling and optimization of classifiers to achieve an optimal energy-latency-accuracy tradeoff. We show through experimentation on five real scenarios, classifiers on Kobe exhibit tight utilization of available resources. For comparable levels of accuracy traditional classifiers, which do not account for resources, suffer between 66% and 176% longer latencies and use between 31% and 330% more energy. From the experience of using Kobe to prototype two new applications, we observe that Kobe enables easier development of mobile sensing and classification apps. © 2011 ACM.