Enhanced Multi-Label Classification Using Fuzzy Deep Neural Networks for Imbalanced Data

Abstract

In classification issues where each data instance is simultaneously assigned many labels, a classification with multiple labels is a useful method for controlling uncertainty. These kinds of circumstances are common in real-world settings where judgments are based on ambiguous or crowded data and flexible categorization techniques are favored. Nonetheless, the issue of class imbalance is a feature shared by several multi-label datasets, where samples and the labels that belong to them are not distributed uniformly throughout the data space. To address the issue of class imbalance, we present a fuzzy logic-based multi-label classification method in this study. One or more labels could be used to convey this categorization challenge. Deep neural networks, which have shown themselves to be incredibly effective in such situations, are used to do away with the necessity for an expert to ascertain the logical principles of inference. The benefits and drawbacks of each strategy can be balanced by integrating deep neural networks with fuzzy inference systems. Accuracy and model application flexibility will be improved by the suggested system, especially when it comes to time limits brought on by causality of reported series of times. The suggested model performs better for continuous data classification than four baseline models, according to tests conducted on a classification with multiple labels dataset about present-day and voltage profiles of various home appliances.