<![CDATA[The need for perimeter security systems is increasingly important in facing the increasing risk of intrusion to various infrastructures. This study aims to compare the performance of the Support Vector Machine (SVM), Random Forest, and Multi-Layers Perceptron Neural Network (MPL-NN) classification algorithms in separating intrusion and non-intrusion data classes recorded in the SW-420 vibration sensor installed on the perimeter fence. The Message Queuing Telemetry Transport (MQTT) communication protocol is used to connect the sensor to the program that records the dataset. Data collection is carried out through simulations of various vibration scenarios, such as intrusion attempts (intrusion) and environmental disturbances (non-intrusion). Normalization and label encoding techniques are applied to help the algorithm read important features in each data point. The results of the study show that of the three algorithms, Random Forest has a higher accuracy value with a value of 97% followed by the MLP-NN Tanh activation algorithm with an accuracy value of 93%. While the SVM algorithm with the RBF kernel has a value of 90.5%. This means that the Random Forest algorithm has good performance in categorizing vibrations.]]>
Copyrights © 2026
