Legacy Versus Algebraic Machine Learning: A Comparative Study

Over the last few decades, researchers have become increasingly interested in machine learning. The field has progressed from classical techniques to neural networks (NNs) and fuzzy neural networks. A novel approach that employs an algebraic model has recently emerged, which enables data conceptualization through generalization and formalization. This parameter-free model has not been shown to suffer from overfitting. This chapter provides an overview of various artificial intelligence methods, including classical methods, fuzzy logic AI, neural networks, continuously constructive neural networks, and neuro-fuzzy networks. The chapter explains the algebraic model in detail and presents it in a simple formal language, rather than using a complex algebraic demonstration. Additionally, the paper compares these approaches qualitatively and quantitatively using the widely used MNIST dataset. This comparison highlights the advantages of the algebraic model over other approaches and illustrates how knowledge propagates through each approach. The research also determines the level of human intervention required.