This study offers a comparison of deep learning models used in fashion recommendation systems, combining sophisticated computer vision methods with Streamlit to create engaging user experience. The research examines four different convolutional neural network (CNN) architectures, which are EfficientNetB0, InceptionV3, VGG16, and ResNet50, to determine which one is the best at extracting features for suggesting clothing items through k-nearest neighbors (KNN). Performance metrics, including accuracy, loss, and validation scores, highlight InceptionV3 as the top-performing architecture, achieving 93.19% accuracy, followed by ResNet50 and VGG16. EfficientNetB0, despite high training accuracy, exhibited severe overfitting and poor validation performance, indicating the need for architectural or training adjustments. The system processes large-scale fashion datasets, leveraging image preprocessing, feature extraction, and vector refinement to enhance recommendation precision. Challenges such as scalability and optimal model selection were addressed through parametric tuning and dataset structuring. Findings demonstrate that InceptionV3 offers superior accuracy, making it the better alternative for fashion recommendation systems. Future research will focus on dataset expansion, real-time processing improvements, and adaptive recommendation mechanisms to further optimize system performance.