Aggregated Machine Learning Algorithm (AMLA) for training Feedforward Neural Network Classifiers

Abstract

Supervised machine learning algorithm is successfully applied in a many real-life classification problems. However, they suffer from the important shortcomings of their high training time which depends on the training dataset size.  Training on very large datasets is a challenging and critical issues nowadays, since the tremendously large amount of data is getting generated every second in many real-time applications. In order to handle this issue, a new learning algorithm called AMLA has been proposed. The proposed AMLA method has been incorporated in the most popular supervised algorithms, Neural Networks, Support Vector Machines and Decision trees, which are extensively used for classification problems in Machine Learning.  It is evaluated effectively using the four benchmark datasets - SPECT Heart, Wisconsin Breast Cancer, Splice-junction Gene Sequences and Drug consumption dataset. Proposed AMLA method proved that it trains faster than existing standard supervised Machine Learning algorithm with improved accuracy rate through simulation.