Simulation and Analysis of Mathematical Methods in Real-Time Engineering Applications
framework based on histogram-gradient feature processing, and deep CNN algorithm with SVM classification is proposed for improving paddy crop cultivation. A deep CNN algorithm is used for noise reduction in unclassified pest images to improve classification under linear SVM. The identification of pest from the de-noised images is performed using a linear SVM classifier along histogram variants embedded with gradient feature. The descriptors feature such as SIFT, SURF, and HOG are computed for all classifiers. It is found that the proposed methodology has evidenced to achieve improved classification when compared with all other existing algorithms.
Chapter 14 describes the term Edge Analytics, which can be defined as tools and algorithms that are deployed in the internal storage of the IoT devices or IoT gateways that collects, processes, and analyses the data at the deployed place itself rather than sending that data to the cloud for analytics. It contains novel end-user application testing equipped with ML on the edge of the IoT devices. A novel framework to achieve this is also proposed. The case study taken is a real-time one and has been tested successfully using the test cases generated on the edge.
Acknowledgments
We deeply indebted to Almighty god for giving this opportunity and it only possible with presents of God.
We extend our deep sense of gratitude to our Son Master H. Jubin, for moral support and encouragement, at all stages, for the successful completion of this Book.
We extend our deep sense of gratitude to our scholars and friends for writing the chapter in time and amenities to complete this book. I sincerely thank our parents and family member for providing necessary support.
We express our sincere thanks to the management of Vellore Institute of Technology, Vellore, India and Anna University, Regional Campus, Tirunelveli. Finally, we would like to take this opportunity to specially thank Wiley Scrivener publisher for his kind help, encouragement and moral support.
—Y. Harold Robinson, Ph.D.
— E. Golden Julie, Ph.D.
I would like to thank the Almighty for giving me enough mental strength and belief in completing this work successfully. I thank my friends and family members for their help and support. I express my sincere thanks to the management of IFET College of Engineering, Tamilnadu, India. I wish to express my deep sense of gratitude and thanks to Wiley Scrivener publisher for their valuable suggestions and encouragement.
—T. Ananth Kumar, Ph.D.
I express my sincere thanks to the management of Vellore Institute of Technology, Vellore, India. Also, I would like to thank the Wiley Scrivener Press for giving me the opportunity to edit this book.
—S. M. Jaisakthi, Ph.D.
1
Certain Investigations on Different Mathematical Models in Machine Learning and Artificial Intelligence
Ms. Akshatha Y* and Dr. S Pravinth Raja†
Dept. of CSE, Presidency University, Bengaluru, Karnataka, India
Abstract
Artificial Intelligence (AI) is as wide as the other branches of computer science, including computational methods, language analysis, programming systems, and hardware systems. Machine learning algorithm has brought greater change in the field of artificial intelligence which has supported the power of human perception in a splendid way. The algorithm has different sections, of which the most common segment is classification. Decision tree, logistic regression, naïve bays algorithm, support vector machine algorithm, boosted tree, random forest and k nearest neighbor algorithm come under the classification of algorithms. The classification process requires some pre-defined method leading the process of choosing train data from the user’s sample data. A host of AI Advanced AI programming languages and methodologies can provide high-level frameworks for implementing numerical models and approaches, resulting in simpler computational mechanics codes, easier to write, and more adaptable. A range of heuristic search, planning, and geometric reasoning algorithms can provide efficient and comprehensive mechanisms for resolving problems such as shape description and transformation, and model representation based on constraints. So behind every algorithm there lies a strong mathematical model, based on conditional probability. This article is the analysis of those mathematical models and logic behind different classification algorithms that allow users to make the training dataset based on which computer can predict the correct performance.
Keywords: Artificial intelligence, classification, computation, machine learning
1.1 Introduction
The increasing popularity of large computing power in recent years, due to the availability of big data and the relevant developments in algorithms, has contributed to an exponential growth in Machine Learning (ML) applications for predictive tasks related to complex systems. In general, by utilizing an appropriate broad dataset of input features coupled to the corresponding predicted outputs, ML automatically constructs a model of the scheme under analysis. Although automatically learning data models is an extremely powerful approach, the generalization capability of ML models can easily be reduced in the case of complex systems dynamics, i.e., the predictions can be incorrect if the model is extended beyond the limits of ML models [1]. A collection of AI ideas and techniques has the potential to influence mathematical modelling study. In particular, information-based systems and environments may include representations and associated problem-solving techniques that can be used in model generation and result analysis to encode domain knowledge and domain-specific strategies for a variety of ill-structured problems. Advanced AI programming languages and methodologies may include high-level frameworks to implement numerical models and solutions, resulting in codes for computational mechanics that are cleaner, easier to write and more adaptable. A variety of heuristic search, scheduling, and geometric reasoning algorithms may provide efficient and comprehensive mechanisms for addressing issues such as shape definition and transformation, and model representation based on constraints. We study knowledge-based expert systems and problem-solving methods briefly before exploring the applications of AI in mathematical modelling.
1.1.1 Knowledge-Based Expert Systems
Knowledge-based systems are about a decade old as a distinctly separate AI research field. Many changes in the emphasis put on different elements of methodology have been seen in this decade of study. Methodological transition is the most characteristic; the emphasis has changed from application areas and implementation instruments to architectures and unifying concepts underlying a range of tasks for problem-solving. The presentation and analysis were at two levels in the early days of knowledge-based systems: 1) the primitive mechanisms of representation (rules, frames, etc.) and their related primitive mechanisms of inference (forward and backward chaining, inheritance, demon firing, etc.), and 2) the definition of the problem.
A level of definition is needed that describes adequately what heuristic programmers do and know, a computational characterization of their competence that is independent of the implementation of both the task domain and the programming language. Recently in the study, many characterizations of generic tasks that exist in a multitude of domains have been described. The kind of information they rely on and their control of problem solving are represented by generic tasks. For expert systems architecture, generic tasks constitute higher-level building blocks. Their characteristics form the basis for the study of the content of the knowledge base (completeness, accuracy, etc.) in order to explain system operations and limitations and to establish advanced tools for acquiring knowledge.