Predicting the Likeliness of Developing Leukemia using Deep Learning and DNA Micro-array Classification Techniques

Abstract

The analysis of DNA Micro-array data is a versatile field of research that is emerging as a mechanism to predict instances of diseases with the aid of a deep neural network. Machine learning-based automated gene data classification is a key component when diagnosing gene-related diseases. The size of data in these problems is generally very large, and thus it is imperative to determine a substantial gene expression classifier that can effectively process the data. To alleviate these issues, deep learning is an advanced machine learning technique used to tackle these roadblocks. A higher number of hidden layers in a deep learning network allows for the processing of this data. The method of classification mentioned in this report allows for the comprehension of the convergence and training of a deep neural network (DNN). This piece of work uses a DNN for the classification of gene expression data. The data-set utilized in this work contains the bone marrow expressions of 72 leukemia patients. To classify the patient as having either acute lymphocyte leukemia (ALL) or acute myeloid leukemia (AML), the network is trained with 80% data while the other 20% is part of the validation set. In this report, the proposed DNN classification model yields an adequate result in comparison to Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Naive Bayes classifiers tested. The two types of leukemia mentioned above are tested yielding 98.2% accuracy, 96.59% sensitivity, and 97.9% specificity with the proposed DNN classification model in this research.