Deep Learning (DL) models, specifically Artificial Neural Networks (ANNs), have surged in popularity due to their brain-like architecture. However, this leads to high energy consumption and demands significant computational resources. To address these issues, researchers are developing new neuromorphic computing systems using electronics and optical components. Optical Neural Networks (ONNs) offer greater computational speed and energy efficiency compared to electronic ones. They utilize the complex light field’s amplitude and phase, providing a richer representation of information. However, traditional Real-valued Neural Networks (RVNNs) can’t fully utilize this property when implemented on optical hardware. In contrast, Complex-valued Neural Networks (CVNNs) excel in information processing but have been underutilized due to the lack of suitable hardware. The emergence of ONNs has revolutionized this. However, ONNs face challenges due to noise in analog hardware, which impacts performance. This study explores the noise-resilience of CVNNs to assess their viability for optical hardware implementation and compares their performance with RVNNs. Noise models experimentally validated are used for testing, and training techniques are suggested to boost performance while computing in noisy conditions.