A Hybrid Approach for Classification and Feature Extraction Using Machine Learning Techniques

Abstract

Artificial intelligence (AI) solutions are used to help make choices that include a high precision of choices they recommend and a deep understanding of choices so that the chiefs can trust them. Verifiable, non-emblematic learning methods have greater perceptual accuracy. Hybrid AI systems analyze the data and exploratory characteristics of approach types. The fundamental purpose of this commitment is to differentiate between a proper AI strategy for choice of assistance that produces reliable and fair outcomes, depending on the various AI techniques, which provide an analysis of different approaches. In this study, we present a hybrid framework for classification and feature extraction. Such a hybrid framework is required for the selection of dataset to Machine learning classifier that will have different results with unlike datasets We have utilized five different types of datasets in this study. As datasets are imbalanced so preprocessing of data is performed firstly. Input with maximum accurate results will be reproduced from our hybrid approach because this approach shows which type of classifiers should be used under what type of dataset, meanwhile exception of the generous fact based on results should be different among different classifiers when applied to a various dataset. After that, a comparative analysis of generic Machine learning algorithms with various datasets has been made as well. The accuracy of the hybrid approaches is compared with the generic approaches, a clear improvement in results in the form of accuracy, precision, recall, and F1 score is found. We used the initial layers of CNN for feature extraction and passess them to ML algorithms for classification.Each information image will go through a progression of convolution layers with channels (Kernels), pooling, fully linked layers (FC) and applying Softmax capabilities to define an object with probabilistic qualities anywhere in the range of 0 and 1. It's one of the simple classes for acknowledging images, arrangements for photos. At the end of this study, we infer which Machine learning algorithm improves the classification accuracy for which type of dataset.