Kasturi Chavan
Computer Science
August 2025
This paper addresses the growing challenge of accurately diagnosing blood cancer using artificial intelligence. By applying supervised models like CNN and Random Forest, and unsupervised models such as K-Means Clustering, we compare their performance on a curated medical imaging dataset. Data augmentation methods like rotation and scaling were applied to enhance model generalization. Among all, CNN with augmentation achieved 95% accuracy. This comparative study suggests machine learning’s practicality in supporting medical diagnostics. Blood cancer, particularly leukemia, remains one of the leading causes of cancer-related deaths globally due to its complex nature and the challenges in early diagnosis. Traditional diagnostic methods, which rely heavily on manual examination of blood smears, are time-consuming, prone to human error, and require expert interpretation. With the growing capabilities of artificial intelligence, especially in medical imaging, machine learning (ML) presents a promising solution to automate and improve the accuracy of blood cancer diagnosis. This research conducts a comprehensive comparative study of both supervised and unsupervised machine learning models for the classification of leukemic and healthy cells using the ALL-IDB image dataset. Models such as Support Vector Machines, Random Forests, and deep learning architectures like ResNetRS50 and EfficientNetB3 are implemented and evaluated. Unsupervised models like k-means and autoencoders are also explored for their utility in anomaly detection. Furthermore, data augmentation techniques, including SMOTE and geometric transformations, are applied to address data imbalance and enhance model performance. The inclusion of explainable AI techniques like LIME and SHAP provides transparency to model predictions, making them more interpretable for clinical use. Our study confirms the potential of ML to assist in early, accurate, and scalable blood cancer diagnosis
340- 352
© Airo Journals 2021
