Hybrid Ensemble Deep Learning Framework for Blood Cancer Identification

Leukemia is a life-threatening malignancy that originates in the blood-forming tissues and rapidly affects the production and morphology of white blood cells. Conventional diagnosis relies on manual inspection of peripheral blood smear (PBS) images by hematologists, a procedure that is time- consuming, subjective, and difficult to scale in real clinical settings. In this work, we present a Hybrid Ensemble Deep Learning framework that combines a transfer-learning based Convolutional Neural Network (CNN), a Multi-Layer Perceptron (MLP), and an Ensemble Voting Classifier to identify blood cancer from microscopic images. The proposed system integrates robust preprocessing, aggressive data augmentation, feature extraction using MobileNetV2, fully connected decision layers and both hard and soft voting schemes. Experiments on a publicly available Kaggle dataset of leukemic and normal smear images achieve an overall accuracy above 95%, with strong precision and F1-score across malignant classes. A graphical user interface implemented using Python Tkin- ter and a Flask web back-end demonstrate that the model can be deployed for real-time, image-based decision support in hospital environments.