Melanoma is a type of skin cancer that benefits from early detection. A non-invasive imaging technique for early detection is dermoscopy. This study investigates the efficacy of various segmentation algorithms and loss functions for the segmentation of skin lesions using deep learning models. This study suggests a Modified UNet(MUnet) model for skin cancer segmentation and assesses its performance on the PH2 dataset. The best results for skin cancer segmentation are obtained using the Modified UNet(MUnet) model along with the AdamW optimizer and Fused loss function. The Fused loss function combines the benefits of binary cross entropy and dice loss for efficient segmentation, while the AdamW optimizer is well recognized for its versatility in deep learning optimization. The study highlights the significance of choosing suitable segmentation algorithms and loss functions for precise skin lesion segmentation. This work achieved an accuracy of 97.42% and loss of 0.000169.