Neural network-based indoor localization system with enhanced virtual access points

Multimodal Approaches for Indoor Localization for Ambient Assisted Living in Smart Homes

“…As outlined in Section 2, one the research challenges in this field of Indoor Localization is the need to develop an optimal machine learning model for Indoor Localization systems, Indoor Positioning Systems, and Location-Based Services. In [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25], researchers have used multiple machine learning approaches-Random Forest, Artificial Neural Network, Decision Tree, Support Vector Machine, k-NN, Gradient Boosted Trees, Deep Learning, and Linear Regression. However, none of these works implemented multiple machine learning models to evaluate and compare the associated performance characteristics to deduce the optimal machine learning approach.…”

“…However, none of these works implemented multiple machine learning models to evaluate and compare the associated performance characteristics to deduce the optimal machine learning approach. Due to the differences in the datasets used or the real-time data that was collected, the associated data preprocessing steps that were different, variations in train and test ratio of the data, and several other dissimilar steps that were associated with the developments of each of these machine learning models as presented in [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25], their final performance accuracies cannot be directly compared to deduce the best approach. Thus, analyzing the performance characteristics of multiple machine learning models, developed, implemented, and tested as per the same methodology, to deduce the optimal approach for development of such Indoor Localization systems serves as the main motivation for the work presented in the section.…”

“…These results are shown in Table 7. In this section, we followed the same steps as outlined in Section 4.3 and as per the flowchart shown in Figure 11, to develop, implement, and test this methodology by using all the machine learning methods that have been used by researchers [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] in this field. These machine learning methods included-Random Forest, Artificial Neural Network, Decision Tree, Support Vector Machine, k-NN, Gradient Boosted Trees, Deep Learning, and Linear Regression.…”

“…Need for an optimal machine learning-based approach for Indoor Localization: A range of machine learning approaches-Random Forest, Artificial Neural Network, Decision Tree, Support Vector Machine, k-NN, Gradient Boosted Trees, Deep Learning, and Linear Regression, have been used by several researchers [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] for development of various types of Indoor Localization systems for IoT-based environments. While each of these systems seem to perform reasonably well but none of these works attempted to develop an optimal machine learning model for Indoor Localization systems.…”

“…Additionally, due to variations in the data source, differences in the types of data, varied methods of data collection, different training set to test set ratios, dissimilar data preprocessing steps, as well as because of differences in the simulated or real-world environments in which these respective systems were developed, implemented, and deployed, the performance metrics of these systems cannot be directly compared to deduce the optimal approach. These works [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] along with the machine learning approaches that were used in each are outlined in Table 17. There is a need to address this research challenge of identifying the optimal machine learning methodology for Indoor Localization.…”