Reevaluating Loss Functions: Enhancing Robustness to Label Noise in Deep Learning Models