PROJECT TITLE :

Boosting Mobile Apps under Imbalanced Sensing Data

ABSTRACT:

Mobile sensing apps have proliferated rapidly over the recent years. Most of them rely on inference elements heavily for detecting fascinating activities or contexts. Existing work implements inference components using traditional models designed for balanced knowledge sets, where the sizes of attention-grabbing (positive) and non-fascinating (negative) knowledge are comparable. Practically, however, the positive and negative sensing data are highly imbalanced. For example, one daily activity such as bicycling or driving usually occupies a little portion of your time, resulting in rare positive instances. Beneath this circumstance, the trained models primarily based on imbalanced data tend to mislabel positive ones as negative. In this paper, we have a tendency to propose a replacement inference framework SLIM primarily based on several machine learning techniques so as to accommodate the imbalanced nature of sensing data. Particularly, guided underneath-sampling is used to obtain balanced labelled subsets, followed by a similarity-based sampling that attracts massive unlabelled knowledge to boost coaching. To the most effective of our information, SLIM is the first model that considers knowledge imbalance in mobile sensing. We have a tendency to prototype two sensing apps and the experimental results show that SLIM achieves higher recall (activity recognition rate) while maintaining the precision compared with 5 classical models. In terms of the recall and precision, SLIM is around percent better than the compared solutions on average.