Prompt detection of malaria is crucial in order to avert outbreaks. Deep learning has exhibited exceptional performance in instances where it comes to identifying tissue images and analyzing cell morphology. Although extensive research has been conducted on the application of deep learning to malaria diagnosis, the majority of these studies have concentrated on binary classification tasks involving structures such as red blood cells and nuclei. This work examines malaria recognition in multiple stages and to introduce the Neighbor Sample Joint Learning (NSJL) model. Neighborhood relationship mining, Convolutional Neural Network (CNN) feature learning, and graph feature integration are all components of NSJL. The process involves the extraction of CNN features, which are subsequently fed into a Graph Convolutional Network (GCN) via a K-nearest neighbor’s adjacency graph. In order to assess NSJL, it is juxtaposed with cutting-edge methodologies. The findings suggest that the NSJL model attains the following performance metrics: 92.50% accuracy, 92.84% precision, 92.50% recall, and 92.52% F1 score. These outcomes illustrate the capability of the method to detect malaria, as its accuracy is at least 7% greater than that of competing approaches.