A Hybrid Principal Label Space Transformation-Based Binary Relevance Support Vector Machine and Q-Learning Algorithm for Multi-label Classification

Classification is one of the most important Machine Learning processes, with numerous applications in a variety of fields. Data are associated with multiple labels in several applications, including text classification, image processing, gene analysis, etc. The problem of predicting the labels of data samples is called multi-label classification (MLC). An effective learning model is required to solve this problem. Recently, scholars have developed a variety of MLC models to solve the most challenging real-life MLC problems. In spite of this, the amount of data, the number of features, and the number of labels are increasing in the big data era, posing critical challenges to existing MLC models. An MLC model is presented to address these challenges and enhance existing models' precision. To encode the raw data in the proposed MLC model, the Principal Label Space Transformation algorithm is used. Using Binary Relevance Support Vector Machine, the initial label set is predicted. To improve initial predictions, a modified Q-Learning algorithm is developed. Using nine real-world datasets, the MLC is compared with seven Commonly used MLC models Considering EM, AP, Macro F1, Hamming Score and Micro F1. MLC model is superior to competitors based on experimental results.