Identification of Lung Cancer Biomarkers from Single-Cell RNA Sequencing Data using Random Forest with Extreme Gradient Boosting

Nowadays, the recent advances in single cell Ribonucleic Acid (RNA) sequencing (scRNA-seq) have allowed the analysis of individual cell transcriptomes for identifying the lung cancer biomarkers. However, Auto Encoders (AE) faced challenges due to the complexity and heterogeneity of scRNA-seq data. Hence, this research proposes Random Forest with Extreme Gradient Boosting (RF-XGBoost) to identify the lung cancer biomarkers from scRNA-seq data for which the data is collected from Gene Expression Omnibus (GEO) dataset. Then, the collected data is pre-processed with Synthetic Minority Oversampling Technique (SMOTE). Further, Boruta model with Minimum Redundancy Maximum Relevance (mRMR) is incorporated as it helps to reduce dimensionality, remove redundant genes and selects features maximizing relevance with target labels while minimizing redundancy respectively. Finally, RF-XGBoost is incorporated as it is effective handling high-dimensional data, reduces overfitting and effective classification. The proposed RF-XGBoost model acquired greater results with an accuracy 95.6% and Matthews Correlation Coefficient of 0.995 when compared to Decision Tree (DT) respectively.