A Hybrid Supervised-Unsupervised Learning Framework for Biomedical Data Analysis and Gene Signature Identification

Abstract

This paper presents a Hybrid Supervised-Unsupervised Learning Framework tailored for Biomedical Data Analysis and Gene Signature Identification. The study meticulously analyzes a dataset comprising patients' information in the biomedical domain, employing rigorous preprocessing steps to ensure data integrity and relevance. Non-contributing attributes such as patients' ages and IDs were excluded from the analysis. Principal Component Analysis (PCA) clustering and hierarchical encoder-decoders were utilized to discern critical genes and evaluate their accuracies within the dataset. To ensure robustness, the dataset was divided into two distinct segments, NO and N1, each comprising 489 rows. Both segments underwent comprehensive analysis, with 208 rows allocated for training data and the remaining rows for testing. A baseline performance metric was established using the Naive Bayes algorithm (Accuracy: 0.85), with 80% of the dataset assigned to training and 20% to testing. Furthermore, the study compares the outcomes of the baseline model with those derived from Support Vector Machine (SVM) (Accuracy: 0.91) and Random Forest algorithms (Accuracy: 0.89), providing insights into their relative efficacy. Through systematic evaluation and comparison of various algorithms, this research contributes valuable insights to the field of biomedical machine learning. The findings have the potential to inform future investigations and aid researchers in making informed choices when selecting appropriate methodologies for analyzing similar datasets.