WiDE WiFi Distance based Group Profiling Via Machine Learning
Abstract : We develop WiDE, a WiFi-distance estimation based group profiling system using LightGBM. Given the uploaded WiFi information by users, WiDE can automatically learn powerful hidden features from the proposed features for between-person distance estimation, and infer group membership with the estimated distance. For each group, WiDE classifies the mobility level, and recognizes the group structure by applying the multi-dimensional scaling technique on the matrix of distance between pairwise persons within the same group.We first validate the performance of between-person distance esti- mation via extensive experiments in a three-floor campus building and a shopping center, showing that WiDE system outperforms other machine learning based approaches for between-person distance estimation, with the average absolute error (AAE) of 0.69m and 1.14m for the campus building and shopping center, respectively, and the corridor identification accuracy for the campus building is over 99%. In addition, the experiments in the shopping center show that our approach can accurately detect groups, classify group mobility into fine-grained level and recognize the group structure.
? The system can enable many existing industries to create new value for their customers by leading prospective buyers to desired products in superstores, allowing students running late to class to more easily navigate their confusing campuses, empowering tourists to find their way around attractions, and helping many others in a multitude of situations. .
? The current implementation differentiates itself from the other various existing ventures because of its independence from a uniquely designed and installed infrastructure - the program runs indoor positioning by sensing and analyzing only the ambient signals of a given location without relying on explicitly placed beacons, or other signal generating objects.
The goal of this stage is to find good initial weights for the different layers of the network, instead of using initial random weights . This is known to speed up the convergence of the training, reduce the possibility of falling in a local minima, and avoid the vanishing gradient problem . In this stage, each autoencoder in the stack is trained independently using the output of the previous autoencoder as its input (input data for the first autoencoder). Despite probabilistic techniques being able to handle the inherently noisy wireless signals in a better way than deterministic techniques, they usually assume that the signals from different APs are independent to avoid the curse of dimensionality problem . This leads to coarse-grained accuracy.
• This system can be improved in further iterations by exploring the benefits of data preprocessing outside merely the realm of simple time-based binning.
• More features can be generated to extract additional useful information from the raw data and reduce the time spent in training; statistical metrics, such as mean, standard deviation, or other Gaussian distribution statistics, can be derived in conjunction with motion data to coordinate dynamic binning intervals, among others.
• By continuing to improve the proposed system, we aim to bring the outdoors in, making indoor navigation as simple and easy to use as GPS is for the outdoors today
? It can also be seen from the figure that WiDeep has the best performance in handling device heterogeneity compared to the other two systems across all percentiles.
? This is due to the combination of additive noise in the training data and the adoption of denoising autoencoders which gives WiDeep greater flexibility than the other systems.
? In particular, this is true since device heterogeneity can be considered to be a form of noise, which the WiDeep network and training process are designed specifically to combat.
? Horus also shows better adaptability than DeepFi.
|