PROJECT TITLE :

Performance Improvement of a Parsimonious Learning Machine Using Metaheuristic Approaches

ABSTRACT:

When dealing with data stream mining, autonomous learning algorithms operate in an online fashion. This is desirable because the least amount of computational complexity possible should be achieved. Because of the relative ease with which they are constructed, parsimonious learning machines (PALMs) are good candidates for use in situations like these. However, the adjustment of the structures of these sparse algorithms, in terms of the addition or deletion of rules, is dependent on previously defined thresholds. In addition, the degree of fuzziness in membership grades is another parameter of PALM that can be adjusted. The most effective combination of these hyper parameters can be derived either from the knowledge of specialists or from optimization strategies, such as greedy algorithms. In this work, a meta heuristic-based optimization technique known as the multimethod-based optimization technique (MOT) is used to develop an advanced PALM. The goal of this work is to reduce the dependency of such experts on computationally expensive greedy algorithms as well as the use of these algorithms by these experts. The effectiveness of several well-known optimization strategies, including the greedy search, the local search, the genetic algorithm (GA), and the particle swarm optimization, has been evaluated and contrasted with the results (PSO). In the majority of tests, the proposed parsimonious learning algorithm with MOT performed better than its competitors. It demonstrates that the multioperator-based optimization technique has advantages over the single operator-based variants when it comes to selecting the best feasible hyperparameters for the autonomous learning algorithm while still maintaining a compact architecture.