A novel leukemic cells feature selection using convolutional neural networks and classification using machine learning algorithms

The application of machine learning and internet of things (IoT) technologies in medical settings has dramatically altered the landscape of leukemic cell research, diagnosis, and treatment. This technological integration has ushered in a new era in the field of leukemia management, significantly enhancing our ability to detect, comprehend, and address malignant cells. This novel approach has transformed the landscape of both leukemia research and patient management. Feature selection of leukemic cells through a classification process using IoT and machine learning represents a great opportunity where both promise and challenges have been faced towards the medical field. This study is aimed at developing a new system to guide the application of technologies for precise identification of leukemic cells. The system employs several feature-selection and classification techniques based on a cell sample analysis of the patients. Because the system for analyzing a large number of medical and biological samples is new and bring the ground baseline for the medical clinicians so that they can discover and develop the algorithms for understanding leukemia and other disease. A hybrid that consists of an SVM algorithm for classifying and CNN model for extracting general features also did brilliantly. It selects features through different machine learning algorithms to identify a minimal feature set that is effective in differentiating high quality cells with high accuracy and precision. A classification model approach was then taken by using list of features to classify cell types by their features. The analysis of neural networks was conducted using the clinically established ALL-IDB, a collection of micrographic blood image sample images that are freely available on the internet. As many existing methods were available, results of the proposed model were compared with gray wolf optimization (GWO), enhanced machine learning (EML) and feature selection-enabled machine learning (FSEML). On 100 samples, the positive predictive value of the proposed model was 95.40%, which is better than GWO (47.82%), EML (89.40%), and FSEML (84.88%).