Abstract:
Localization is a topic widely discussed in the field of networking. It can be classified
into indoor and outdoor localization, with indoor localization being particularly
significant for the development of smart cities. Unlike outdoor localization, which
commonly relies on GPS, indoor localization faces challenges due to obstacles such as
walls. Consequently, there is a need for a novel indoor localization model, which forms
the main focus of our research. The objective of this study was to apply an accurate deep machine learning hybrid model for predicting smartphone locations in indoor
environments. To accomplish this, we conducted a comparative analysis of performance
accuracy among machine learning algorithms (SVM and KNN) and recurrent neural
network algorithms (Bi LSTM and LSTM). Our findings revealed that SVM consistently
outperformed KNN across all smartphones, while Bi LSTM exhibited superior
performance compared to LSTM. Building upon these results, we devised a collaborative
multi-stage structure for our hybrid model, combining the strengths of Bi LSTM and
SVM, resulting in the Bi LSTM-SVM model. For our study, we gathered a dataset by
utilizing diverse smartphones and access point models. These smartphones collected
received signal strength from the access points, which was then adjusted to create an
indoor localization dataset. The dataset has passed different preprocessing steps to ensure
its compatibility with the models employed. We evaluated the performance of five
models: SVM, KNN, Bi LSTM, LSTM, and our proposed Bi LSTM-SVM hybrid model,
with our proposed model demonstrating superior performance. Furthermore, we
investigated the impact of human presence on indoor localization accuracy, as well as the
relationship between the number of access points and localization accuracy. Each
smartphone exhibited varying levels of accuracy across the five models. To reconcile
these differences, we employed calibration techniques. Consequently, our proposed
hybrid model, Bi LSTM-SVM, achieved an accuracy of 89.11% in scenarios without
human presence and 78.7% in scenarios with human presence.
Keywords: Localization, Indoor Localization, Outdoor Localization, Hybrid Model,
Receiving Signal Strength