Towards temporal associative classification using triclustering

Abstract

This thesis presents the development of a temporal associative classifier aimed at addressing the inherent complexities of biomedical data analysis through the use of discriminative triclusters. Biomedical datasets are often characterized by high dimensionality, temporal misalignment, and missing data, which pose significant challenges for traditional classification methods. The developed classifier leverages triclustering techniques to uncover patterns within threedimensional substructures, particularly focusing on the temporal dimension. These patterns, identified by triclustering algorithms, are utilized to enhance the accuracy and interpretability of predictions within clinical contexts. The research involves a comprehensive evaluation of the classifier’s performance against established Triclustering-Based approaches using Random Forest and XGBoost classifiers, which are also designed to handle three-way temporal data. The evaluation metrics include accuracy, specificity, and sensitivity, carefully chosen to reflect the clinical importance of minimizing both false positives and false negatives. The results indicate that the associative classifier exhibits competitive performance, particularly in scenarios where constant, additive, and multiplicative patterns are predominantly present in the context dimension. However, the study also reveals limitations, notably the classifier’s dependency on the quality and quantity of triclusters generated, which significantly influences its overall effectiveness. Despite these limitations, the classifier demonstrates potential in condensing complex triclusters into a more interpretable set of rules, thereby enhancing its applicability in clinical settings where both accuracy and interpretability are critical. The study contributes to the field of temporal data classification by providing valuable insights into the use of triclustering in biomedical data analysis. It also provide the basis for future research aimed at refining triclustering algorithms, and expanding the developed classifier’s capability to capture patterns on the feature dimension. Ultimately, this research aspires to advance precision medicine by offering a robust tool that can improve diagnostic accuracy and treatment personalization, thereby positively impacting healthcare outcomes.