Multivariate or univariate model analysis for indoor location systems: A comparison

Using a set of signal features (e.g. magnetic-field, light, sound), we can create a model to estimate the location of a user in an indoor environment. In our previous work, we proposed a model with 46 features from the temporal and spectral evolution of the magnetic-field, which enables us to estimate the indoor location using the readings of a magnetometer built in a smartphone. Considering our previous results, some questions arise. Firstly, (1) Is it possible to estimate the location of a user in indoor environments with a single feature of the magnetic-field signal (univariate model)?; (2) Which model, univariate or multivariate, has higher performance in terms of accuracy? and (3) Which type of features (frequency/temporal) outweigh the indoor location? To answer these questions we propose an evaluation which consists of four experiments: i) building a model per feature, ii) constructing a model that contains only spectral features; iii) building a model composed by the temporal features, and iv) create a model that combines all features. Our results indicate that a multivariate model has higher accuracy in comparison to the univariate models, however, in the multivariate model a clever feature selection is needed to reduce the number of features required to explain the variance of our multivariate model.