Discovering Leukemia Insights: Comparing Traditional and Modern ML Techniques for Blood Cancer Diagnosis

Abstract

Leukemia, a lethal cancer of the blood, needs to be diagnosed in a timely and accurate manner in order to improve patient survival rates, yet standard methods take too much time and consume too many resources. We investigate, in this study, how machine learning (ML) improves leukemia diagnosis by combining traditional and modern model efficiencies to analyze blood sample data with accuracy and speed. Our methodology begins with preprocessing blood samples — microscopic images or clinical readings — and then splitting the data into test and training sets to ensure robust testing. We train and test a range of models: Random Forest (RF), which is understandable when dealing with structured data, and deep learning models VGG16, VGG19, EfficientNetB3, and EfficientNetB5 that excel at detecting extremely complicated patterns in cell images. By evaluating such models on criteria like accuracy, sensitivity, and computational efficiency, we aim to identify tools that are not only accurate but also pragmatically deployable in real-world clinical settings, even with limited resources. To enable trust, we provide explainability tools like SHAP and gradient-weighted class activation mapping, which enable clinicians to comprehend the model's predictions. Our findings are a testament to the complementarity of the models: RF works best in low-data settings, while EfficientNet models deliver best-in-class results on image-dominant datasets, paving the way toward rapid, precise leukemia diagnosis. This work points to the promise of ML to get around data imbalance and computational constraints, paving the way toward cheap and trustworthy diagnostic tools for leukemia.

Received: 30 November 2025 | Revised: 18 March 2026 | Accepted: 15 April 2026

Conflicts of Interest

The author declares that he has no conflicts of interest to this work.

Author Contribution Statement

Aseel Alshoraihy: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization.