Leukemia is a malignancy that originates in the bone marrows and it is characterised by aberrant white blood cell growth. Artificial intelligence has flourished in recent years across all scientific disciplines. The accuracy of predicting the initial severity of this infection using artificial intelligence in medical research has increased. The proposed model uses transfer leaning approach with VGG-19, and ResNet-50. The input images are pre-processed by weighted distribution and gamma correction techniques; from this the edges are detected by the Sobel edge detector. The structural features are extracted by the deep neural networks and acquired as feature sets. These feature sets are fused by least absolute shrinkage and selection operator (LASSO) and the support vector machine (SVM) is utilized to categorize the four types of leukemia and healthy. When compared to the results produced by the existing deep neural networks, the proposed approach produces the most precise and effective outcomes. This model yields the accuracy rate of 99.08% and 99.02% for the classification of leukemia.