A novel Hybrid Exhaustive Search and data preparation technique with multi-objective Discrete Hopfield Neural Network

Abstract

The primary objective in building predictive analytics models is to achieve optimal accuracy with real datasets. The limitations of existing models lie in their storage capacity, which hinders the progress of generating high accuracy. When a model storage capacity is limited, it may struggle to process large datasets and encounter underfitting issues, preventing it from capturing the complexities of the data. Hence, this paper addresses these challenges by introducing a novel approach to predictive analytics, focusing on expanding the storage capacity of the Discrete Hopfield Neural Network (DHNN). First, this paper employs satisfiability logic to represent the attributes of a dataset in the DHNN. This logic representation enables the model to establish a connection between neurons and attributes, enabling efficient information processing. Second, to introduce a multi-objective DHNN, a key innovation that enhances the model’s storage capacity. In this context, a novel training algorithm named Hybrid Exhaustive Search is developed to optimize the DHNN’s training phase. Third, this paper introduces new data preparation techniques, including feature selection and a method for identifying the best-induced logic. This best-induced logic explains and extracts dataset relationships. Finally, the proposed model is evaluated based on four reputable metrics and a variety of datasets primarily collected from the UCI Repository. The performance of the proposed model is compared with three existing models. Through extensive experiments and rigorous evaluation, the proposed model can outperform existing models in all metrics demonstrating the effectiveness of expanded storage capacity and novel data preparation techniques employed.