In their study, the researchers introduce a binary classifier titled WiFiBoost based on AdaBoost (Adaptive Boosting) and designed for mobile devices. The boost relies on signal strength indicators (RSSIs) received from WiFi beacons of the existing access points.
The authors describe the learner by stating that it is based on the feature of relative comparison between measured APs in order to solve the problem of device heterogeneity. As for the algorithm, it maintained a set of weights over the training set: at first all weights were set equally and later on each round the weights of wrongly classified examples were increased in order to make a weak learner focus on hard observations.
The testing of WiFi Boost took place in the university campus and included 1,735 observations (890 indoor and 845 outdoor) with the use of 135 access points (APs) and different mobile devices. The results are quite positive: when three and five learners are added per stage, there were 49 and 64 different APs in the final classifier respectively.
As a result, Canovas, Lopez-de-Teruel and Ruiz conclude that “WiFiBoost is an efficient and fast binary classifier which combines several weak learners mainly based on the feature of relative comparisons between measured pairs of access points.” Finally, it is worth noting that the proposal was compared with other techniques such as nearest neighbor (NN) and naïve Bayes (NB) demonstrating that good performance was delivered even with a more compact classification model.
Interested in the paper? Read the full version here