@CONFERENCE\{IMM2010-05915,
    author       = "B. S. Jensen and L. K. Hansen and J. Larsen and J. E. Larsen and K. Jensen",
    title        = "Predictability of Mobile Phone Associations",
    year         = "2010",
    month        = "sep",
    pages        = "91-105",
    booktitle    = "21st European Conference on Machine Learning",
    volume       = "",
    series       = "",
    editor       = "Mining Ubiquitous and Social Environments Workshop (MUSE2010)",
    publisher    = "",
    organization = "",
    address      = "",
    url          = "http://www.kde.cs.uni-kassel.de/ws/muse2010/proceedings.pdf#page=91",
    abstract     = "Prediction and understanding of human behavior is of high importance in many modern applications and research areas ranging from context-aware services, wireless resource allocation to social sciences. In this study we collect a novel dataset using standard mobile phones and analyze how the predictability of mobile sensors, acting as proxies for humans, change with time scale and sensor type such as {GSM} and {WLAN}. Applying recent information theoretic methods, it is demonstrated that an upper bound on predictability is relatively high for all sensors given the complete history (typically above 90\%). The relation between time scale and the predictability bound is examined for {GSM} and {WLAN} sensors, and both are found to have predictable and non-trivial behavior even on quite short time scales. The analysis provides valuable insight into aspects such as time scale and spatial quantization, state representation, and general behavior. This is of vital interest in the development of context-aware services which rely on forecasting based on mobile phone sensors."
}