Learning a multi-criteria classification method using machine learning & metaheuristics techniques

Data classification is a widely used approach in the area of data mining. Methodologies for addressing data classification have been developed in a variety of research disciplines, including artificial intelligence (AI) and Multi-Criteria Decision Aid (MCDA). The objective of this thesis is to develop a new framework for learning the MCDA method PROAFTN. The limitations of PROAFTN are largely due to the set of parameters required to be obtained to perform the classification procedure. That is, to apply PROAFTN, the values of several parameters need to be determined prior to classification, such as boundaries of intervals and weights. In an MCDA context, these parameters are usually dependent on the judgment of the decision maker (DM). This approach has shortcomings, such as being time consuming and dependent on the availability of a qualified DM. To overcome these limitations and to obtain the best parameters from data, an automatic approach is proposed in this work . This thesis introduces new methodologies based on using machine learning and metaheuristic techniques for establishing PROAFTN parameters from data during the training process. The goal is to obtain from training data the best PROAFTN parameters that achieve the highest classification accuracy. To achieve this, different learning methodologies are proposed in this thesis. Firstly, discretization techniques and an inductive approach are introduced to obtain the required parameters for PROAFTN. Secondly, a different approach based on metaheuristic/hybrid-metaheuristic algorithms is used to develop PROAFTN parameters. The use of metaheuristics to learn PROAFTN begins with the formulation of the optimization problem. Then, population-based methods, namely Particle Swarm Optimization (PSO) and Differential Evolution (DE), and the single-point search method Reduced Variable Neighborhood Search (RUNS) are utilized to obtain the best PROAFTN parameters that can be applied on unseen datasets. To test the performance of the proposed learning approaches, their effectiveness in classification is evaluated on several public-domain datasets and compared to a number of well-known machine learning classifiers. Advanced statistical tests such as the Friedman and Nemenyi tests are used for more meaningful comparisons. The general comparative study, including computational results, demonstrates that the proposed approaches are very competitive with and outrank widely used classification algorithms.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.