The Theory of Probabilistic Hierarchical Supervised Learning for Classification
Ziauddin Ursani · 2023
The supervised learning methods for classification are being extensively used in many application areas including health, education, industry, agriculture, and defense. Therefore, any contribution in this area has potential of application over several aspects of modern life. The theory of probabilistic hierarchical supervised learning for classification has evolved through the years with several publications to its credit. The theory introduces the problem decomposition strategy for training of classification models. The training set is decomposed into smaller hierarchical subproblems, and the model is trained for each subproblem separately. Therefore, multiple trained models are put together in a hierarchical order. The hierarchical model then can be used to classify the samples in the test set. In this paper, further enhancements are introduced to the theory like incorporation of additional mathematical operators to the models, design of hierarchical multimodal fitness function and modified data normalization scheme. In addition, basic principles of theory are further modified and redescribed with greater detail here. These modifications have enabled application of the theory to more datasets. The results are competitive with the well-known methods.