Improved no-line-of-sight static and dynamic sensor data classification using KNN algorithm with PLS model

Abstract

The rising cases of structural collapses across the world have aggravated the problem of finding people under rubble as part of search and rescue (SAR) effort. The conventional search techniques, namely drill operation and the use of dog searching, are usually slow, labour-intensive and unsuccessful in difficult debris setting. Radar systems, though non-invasive, are limited by attenuation and multipath interference in non-line-of-sight (NLOS) environments. The proposed research will contribute to the improvement of victim recognition by creating an advanced machine learning (ML) model that will operate in the most challenging environmental settings. It suggests a modular prediction model combining both the K-nearest neighbor (KNN) and partial least squares (PLS) to extract features and reduce the dimensions. The procedure includes the derivation of essential signal characteristics, dataset validation, PLS application to get limited and discriminative feature amounts, and KNN classification under conditions of both fixed and dynamic conditions. Experimental findings indicate classification scores of 87.87 and 75.70 respectively in case of static and dynamic data. These results validate the practicability of the suggested solution in enhancing the forecasting accuracy during NLOS circumstances and emphasize its possible use in enhancing quicker, more dependable, and evidence-based official choices during the actual SAR operations.