An efficient class-dependent learning label approach using feature selection to improve multi-label classification algorithms

Recently, multi-label categorization learning has emerged as a new area of study in machine learning since it offers a multi-dimensional perspective of the multi-dimensional object. To build a multi-label classification (MLC) system that is both quick and efficient, one must consider the large data environment. This paper introduces a new MLC model, RR-IPLST (Ridge Regression-Principal Label Space Transformation), which uses a combination of ridge regression and original label space transformation (IPLST). The model includes Advanced Orthogonal Projection to Latent Structures (AOPLS) for optimal dimensionality reduction, increasing data set relevance. IPLST uses Singular Value Decomposition (SVD) to identify label correlations, while RR mitigates multicollinearity problems. Unsupervised learning, especially the value learning method, refines label predictions. The proposed model is rigorously evaluated on ten diverse multi-label datasets and demonstrates its effectiveness. Support vector machine, simple Bayes polynomial, k nearest neighbor, decision trees, linear and discriminant analysis are five regularly used data set classification techniques whose outputs are compared. The simulation results showed that the application of the proposed solution is of higher quality compared to similar methods in terms of accuracy, precision, convergence, error measurement, stability and other tested parameters in different data sets. So that the evaluation and test results of the proposed method show a 92.21% improvement compared to other existing algorithms.