Advanced genetic algorithm (GA)-independent component analysis (ICA) ensemble model for predicting trapped humans through hybrid dimensionality reduction

Abstract

The collapse of man-made structures, which frequently bury people beneath rubble, is one of the main causes of death worldwide. Natural disasters and human inefficiency, degradation, and decay are the main causes of this. So, in these urgent situations, a quick and effective deployment is crucial. No-line-of-sight (NLOS) signal analysis, which is the fundamental of seeing through the wall of critical information that is not visible, is disclosed through analysis, and developmental systems can be seen for identifying imprisoned human victims and detecting differences in disaster scenarios. This technology is essential for locating stranded people and evaluating different crises. Ultra-wideband (UWB) signal data, augmented by machine learning techniques, provides a large and quantified output that is useful for applications including engineering, scientific research, and Search and Rescue (SAR) operations. However, this method's primary drawbacks are its large dimensionality, infrequency, and noise, which makes catastrophic scenario prediction difficult. The curse of dimensionality has been addressed in a variety of ways, but issues with accuracy, dependability, and scalability still exist. To choose relevant subset features from the data and for better generalization in various contexts, this work uses an adaptive human presence detector algorithm that hybridizes dimensionality reduction techniques genetic algorithm (GA), which maximizes feature selection, and independent component analysis (ICA), which lowers the dimensionality of the chosen features. The features are sent into the classifiers for technique analysis based on their class variants. A bagged ensemble machine learning classifier was used to assess the reduced dataset, and among other performance measures, The findings with accuracy are 85.69 %, sensitivity of 79.30, and specificity of 91.67 %. These outcomes show how the suggested hybrid strategy can be used to efficiently pick and classify features in NLOS signal data processing. According to the results, this approach may improve catastrophe scenario prediction and support faster and more precise search and rescue efforts by supplementing and enhancing current machine-learning techniques. The findings suggest that this method could complement and enhance existing machine learning techniques, improving disaster scenario prediction and aiding in more accurate and timely search and rescue operations.