Paper Title
RADIANT- Radar AI for Intelligent Automotive & Defence Target Recognition

Abstract
This paper compares different machine learning and deep learning models for radar-based target classification using Radar Cross Section (RCS) signatures under both clear and occluded conditions. The study utilizes simulated radar data along with real-world automotive radar datasets to classify multiple target types such as vehicles, pedestrians, and other objects, while explicitly modeling the impact of occlusion on radar returns. The performance of classification algorithms including Logistic Regression, Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Random Forest, Decision Trees, and Convolutional Neural Networks (CNN) is evaluated using key metrics such as accuracy, precision, recall, F1-score, Matthews Correlation Coefficient, and prediction coverage. In addition, a confidence-based rejection threshold is incorporated to identify and reject low-confidence predictions, thereby reducing misclassification risk in safety-critical scenarios. The experimental results indicate that deep learning models achieve superior robustness and classification performance under occlusion, while SVM and Logistic Regression demonstrate competitive accuracy for limited datasets. Random Forest and kNN show moderate performance, whereas Decision Trees exhibit comparatively lower accuracy. The results highlight the effectiveness of combining RCS-driven learning, occlusion awareness, and rejection-based decision control for reliable radar-based target classification in surveillance and defense applications. Keywords - Radar Cross Section, Target Classification, Occlusion Phenomenon, Rejection Threshold, Radar Signal Processing, Machine Learning, Deep Learning, SVM, CNN