The classification of skin lesions is a crucial process engaged in the early detection and diagnosis of conditions such as melanoma. In this study, we employ fusion method and GRU model to improve VGG16 and HOG based features extraction for lesion classification into bening or malignant categories by extracting sequential, global, and local image features, respectively. To enhance classification, various combination of features like GRU + VGG-16, GRU + HOG, and GRU + VGG16 + HOG are explored. We apply Dimension reduction by Principal Component Analysis (PCA) and subsequently use Support Vector Machines (SVM) as the first-level classification algorithm. GRU + VGG16 + HOG produced the highest result with a percentage of accuracy 87.34%, which was followed by GRU + VGG16 (84.11%) and GRU + HOG (79,20%), which were less accurate. These outcomes demonstrate the robustness of GRU in improving the classification accuracy through learning sequential features, particularly when combined with additional feature extraction techniques. Moreover, the combination method, dimensionality reduction, and utilization of SVM help in developing a very high accuracy and efficient classification system.