Optimization Enabled Ensemble Learning for Leukemia Classification Using Microarray Data

Leukemia classification involves identification and categorization of various leukemias, a cluster of blood malignancies influencing white blood cells. Proper classification is crucial for selecting the appropriate treatment modalities and predicting outcomes in patients. Historically, leukemia classification was based on clinical and morphological characteristics, but new developments in genomics like microarray and next-generation sequencing tools have facilitated more accurate molecular classifications. Machine learning (ML) and deep learning (DL) methods have transformed leukemia classification by enabling automation of analysis in large and intricate datasets to ensure more accurate and efficient leukemia subtype classification. The primary goal of this research is to suggest a new leukemia classification method using microarray data. Leukemia microarray data first undergoes preprocessing, after which feature selection is performed through Serial ExponentialSecretary Bird Optimization Algorithm (SE-SBOA). SE-SBOA is an optimization method that embeds the exponential weighted moving average concept (EWMA) into Secretary Bird Optimization Algorithm (SBOA). The method helps to find the best feature subset, improving model performance at lower complexity. Lastly, leukemia classification is done using the proposed ensemble method that combines Graph Neural Network (GNN), Multi-Layer Perceptron (MLP) and Random Forest. Utilizing the advantages of GNN, MLP and Random Forest, the model proposed herein attains higher classification accuracy and proves to outperform traditional methods. Experimental results demonstrate that the SE-SBOA-based Ensemble Learning technique outperformed standard methods, attaining an accuracy of 95.9%, a precision of 96.1%, a recall of 96.2%, and an F1-score of 96.2%.