Applying random linear oracles with fuzzy classifier ensembles onWiFi indoor localization problem

People localization is required for many novel applications such as proactive caring for the elders or people suffering degenerative dementia. In a previous contribution, we introduced a system for people localization in indoor environments based on a topology-based WiFi signal strength fingerprint approach. The well-known curse of dimensionality critically emerges when dealingwith these kinds of complex environments. We address the localization task as a high dimensional classification problem that can only be effectively addressed by an advanced classifier ensemble approach. Therefore, in this paper we present a localization system based on a fuzzy rule-based classifier ensemble framework where we consider a random linear oracle for the component classifier generation, as this fast and generic method induces more diversity thus improving the final performance. The proposed system is validated in a real environment, achieving very promising results. Its ability to handle the huge uncertainty that is characteristic of WiFi signals is demonstrated.

keywords: Bagging, Classifier ensembles, Fuzzy rule-based classifier ensembles, Random linear oracles, Random subspace, WiFi localization