Outlier Detection in Indoor Localization using "Random Forest" and "Support Vector Machine"

• Published in: 2023 Intelligent Computing and Control for Engineering and Business Systems (ICCEBS) pp. 1 - 5
• Main Authors: Uganya, G., Sathesh, M., Sheela, M. Sahaya, Nalini, N.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14-12-2023
• Subjects: Innovative Indoor localization; Location awareness; Machine learning; Machine learning algorithms; Outlier Detection; Radio frequency; Random Forest; Random forests; Statistical analysis; Support Vector Machine; Support vector machines; Vectors
• DOI: 10.1109/ICCEBS58601.2023.10449262

The objective of this work is to detect outliers in indoor localization using machine learning such as "Random Forest" ("RF") and "Support Vector Machine" ("SVM"). These two algorithms are considered as two groups. Group1 is "RF" by taking 20 samples and group2 is "SVM" by taking 20 samples that are analyzed with 80% of pretest power and 0.05 alpha value. Random Forest Algorithm achieves an accuracy of 83.9%, precision rate of 83.23% and recall rate of 83.77% in a balanced dataset. In an unbalanced dataset, it achieves an accuracy of 64.9%, precision rate of 64.03% and recall rate of 65.48%. Support Vector Machine achieves an accuracy of 82.9%, precision rate of 82.08% and recall rate of 82.41% in a balanced dataset. In an unbalanced dataset, it achieves an accuracy of 59.76%, precision rate of 60.36% and recall rate of 60.72%. The p value of 0.02 for accuracy, 0.03 for precision and 0.01 for recall rate was obtained ("p < 0.05"). From statistical analysis, "RF" achieves significantly better accuracy, precision and recall while compared to "SVM".