Detecting Indoor/Outdoor Places Using WiFi Signals and AdaBoost

One of the key points in any localization proposal for indoor environments is to determine in a precise way whether a device is inside or outside a particular building. That preliminary step of binary classification is required, for example, in order to send sensor readings to a location-based service. In this paper, we present a binary classifier, based on the well-known machine learning meta-algorithm AdaBoost, which makes use of the received Wi-Fi signals in order to infer the indoor/outdoor condition. The simple, so-called, weak learners, which lie at the heart of every AdaBoost implementation, are based in our case in simple RSSI magnitude comparisons between signal observations. Our solution requires only a coarse-grained training stage to produce two sets of samples (indoor and outdoor) that will be used to calculate the final model composed by several of these extremely simple weak learners. The resulting model is therefore really compact and efficient, which provides scalability to calculate in a fast way whether a device is inside or outside a particular building. In addition, we propose a slight modification of the base AdaBoost meta-algorithm to minimize the negative effect of unpredicted AP failures or changes in the network infrastructure, which will result in a moderate impact on the classification error under those circumstances. In order to verify empirically our proposal, we conducted several tests and we compared our results with those obtained using other well-known classification techniques. Our experiments made use of several devices in two different buildings and show that the resulting performance, a mean error rate around 2.5%, is satisfying. We also depict some applications of our system for several scenarios.