Detection of acute lymphocytic and myeloid leukaemia with optimized dimensionality reduction and deep learning classification framework.

Leukaemia is a deadly illness that affects a huge number of people all over the world. ALL (Acute lymphocytic Leukaemia) and AML (Acute Myelogenous Leukaemia) are the common type of leukaemia where AML widely affects older people and ALL commonly affects children. Therefore, the projected system employed GMLP-MLDA (Globalized Multi-Layer Perceptron with Modified Linear Discriminant Analysis) for the detection of ALL and AML through the gene expression dataset. Although the LDA is a better system for reducing dimension, it also has limitations, like improper handling of high-dimensional data, lesser samples, and higher feature values. To optimize classical LDA, the proposed system utilized the optimal fisher criterion with covariance. The classification in the projected system was performed by the MLP (Multi-Layer Perceptron) for its capability of handling non-linear complex relationships among the input and output. To enhance the performance of MLP, the projected system used global feature relation samples. The projected system utilized a gene expression dataset that comprises data on the gene of the leukaemia-affected patients and analyses the leukaemia. The performance of the proposed system is compared internally with three classifiers such as DT (Decision Tree), KNN (K-Nearest Neighbour), and XGBoost. Moreover, comparison of the traditional LDA with the modified LDA is performed to expose the efficiency of the proposed dimension reduction system. Significantly, the proposed system has achieved an accuracy of 95% and demonstrates the model's ability to contribute the fields of medical science and molecular biology, with enhanced leukaemia diagnosis for improving the survival rate.