Modeling Learning Automata based classifification algorithms

Learning Automata is extensively used as a tool by researchers to achieve solutions to various problems pertaining to engineering. One such application is a learning automata based classi- fication algorithm. However, at present, learning automata based classifiers have been limited to only 2-class classification. In this research two learning automata based models, for multi- class classification, have been presented. The performance of the proposed techniques has been evaluated using data sets that are extensively used for benchmarking purposes. Since the ultimate goal of an efficient model is to be accurate with low computational complexity, both the performance and complexity of the proposed techniques have been evaluated us- ing the benchmark data sets. Furthermore, the performance of the proposed model has been compared to the performance of existing and extensively used machine learning algorithms on the same benchmark data sets. Apart from modeling and testing the algorithms, the present research also proposes a methodology for evaluating the performance of machine learning algorithms. A detailed study of the existing performance evaluation techniques for machine learning algorithms has been provided; highlighting their drawbacks and then proposing a simple and efficient technique for evaluating their performance. For both pieces of the work, i.e. both the proposed learning automata based multi-class classification algorithms and the performance evaluation methodology, the theoretical analysis has been carried out in a tight mathematical framework supported by simulation results. A theoretical background of learn- ing automata has also been provided, along with a description of the evolution of the proposed algorithms from the basics of learning automata theory.