Comparison of classifiers for use case detection using onboard smartphone sensors

Onboard smartphone sensors provide ample data modalities which can be used to determine the way a phone is being used. However, in order for use case detection systems to be unobtrusive to users, the classification algorithms and the number of sensors should be kept simple and at a minimum. In this paper light, accelerometer and orientation sensor measurements are recorded for 4 different phone use cases and results from 3 different classifiers (K-means, Naive-Bayes, Neural Network) are compared to identify the sensor modality and classification algorithm that provides the highest accuracy for use case detection. The onboard accelerometer is found to be the sensor modality with highest accuracy across all the classifiers, and the neural network is identified as being the best performing classifier. A discussion of the results linking back to theoretical aspects of the classifiers is also given.

Automated summary

This paper investigates the use of smartphone sensors for detecting phone use cases. By comparing results from different classifiers, it is found that the onboard accelerometer provides the highest accuracy as a sensor modality, and neural network performs the best as a classifier. The paper also includes a discussion on the theoretical aspects of the classifiers.