Crowdsensing-based WiFi Indoor Localization using Feed-forward Multilayer Perceptron Regressor

Most RSS based indoor localization algorithms require the a priori knowledge of location of Access Points, timewise variation of location of user, and use of multiple sensor data. The paper proposes an innovative approach combining the Crowdsensing based wireless indoor localization technology with Artificial Neural Networks, to automatically predict new users location and analyze the effect of device heterogeneity on the RSS localization accuracy, by using cell phone user data. The performance evaluation demonstrates that the trained MLP Regression model can obtain the highest localization accuracy than the probabilistic localization algorithms, without individual model for each device in the fingerprinting database. In contrast with existing systems proposed in the literature, the result shows that our proposed approach efficiently handles very large number of Access Points in 10 times larger indoor spaces.