Machine Learning for Decision Sciences with Case Studies in Python

S. Sumathi, Suresh Rajappa, L Ashok Kumar, Surekha Paneerselvam

• Publisher: CRC Press
• Year: 2022
• Print ISBN: 9781032193564
• eBook ISBN: 9781000590975
• Topic: Computer Science
• Subtopic: Programming Languages
• Index: Computer Science

This book provides a detailed description of machine learning algorithms in data analytics, data science life cycle, Python for machine learning, linear regression, logistic regression, and so forth. It addresses the concepts of machine learning in a practical sense providing complete code and implementation for real-world examples in electrical, oil and gas, e-commerce, and hi-tech industries. The focus is on Python programming for machine learning and patterns involved in decision science for handling data.

Features:

• Explains the basic concepts of Python and its role in machine learning
• Provides comprehensive coverage of feature engineering including real-time case studies
• Perceives the structural patterns with reference to data science and statistics and analytics
• Includes machine learning-based structured exercises
• Appreciates different algorithmic concepts of machine learning including unsupervised, supervised, and reinforcement learning

This book is aimed at researchers, professionals, and graduate students in data science, machine learning, computer science, and electrical and computer engineering.

1 Introduction

DOI: 10.1201/9781003258803-1

Learning Objectives

At the end of this chapter, the reader will be able to

• Understand the need for data science and machine learning (ML) in a real-world scenario.
• Appreciate the fundamentals of ML and deep learning.
• Describe structural patterns and understand their variants.
• Know the key roles of a data scientist for solving practical problems.
• Identify the real-world application areas of ML models.

1.1 Introduction to Data Science

The process of considering the algorithm and coding it with the help of a programming language can be executed by the computer. IT world has made people’s life easier by providing helpful software. A design pattern provides a general refusal solution to the common problems that usually occur in software design. The patterns identify the connections between classes and objects. The primary purpose of patterns is to boost the performance of the development process. This chapter will understand the need for data science and machine learning (ML) in real-world scenarios, highlight the basics of ML and deep learning (DL), know how to describe structural patterns, and understand their variants. In addition, a detailed description of the key role of a data scientist in solving real-world applications is highlighted. Several areas in which ML algorithms are applied are also illustrated.

Data science is a part of software engineering in which we study where the information comes from and its organization. It represents how we can turn the valuable resources in the initialization of any business and the information technology strategies. Data science allows mining a large amount of data that helps to identify the patterns for any organization’s efficiencies. There are some fields of data science: statistics, disciplines of computer science, data visualization, and data mining. Data scientists possess a combination of ML and analytic and statistical skills and experience in coding and algorithms. Data science is everywhere in today’s world. Let’s imagine that you are traveling to a new city. The user can search for some tourist places, restaurants, parks, hotels, etc., with just one click. How is this possible? It’s the data science that saves and displays the results on your phone screen in a few seconds.

1.1.1 Mathematics

Mathematics is everywhere around us, from counting money, temperature calculation, shapes, etc. Data science and mathematics have a strong relation in all the decision-making processes like predicting the routes and searching for the best possible answer. All these tasks are incomplete without mathematics. Data science uses ML algorithms for decision-making processes.

1.1.2 Statistics

Data science is nothing without statistics, and data scientists require a graphical representation of the data with less theory. Business providers prefer data visualization like a bar chart as it becomes easier to evaluate the data and gain more information.

1.2 Describing Structural Patterns

Creating an application or software is a complex task. It becomes more difficult if not created with proper flow or patterns – software engineers use some patterns to symbolize their application patterns to solve many problems. The selection of patterns depends upon the application requirements. Design patterns boost the development process; structural patterns are all about the classes and their instances creation. For example, to create a website that stores student records, the class will be Student and will have attributes like student name, age, father’s name, and address.

1.2.1 Uses of Structural Patterns

Structural patterns play an essential role in effectively solving the problems, as it covers the central concept of object-oriented programming, i.e., classes and their inheritance. These patterns allow classes to work together as a group. With the help of structural patterns, software engineers write less source code as most of the functions are extracted from inherited classes. Some of the commonly used design patterns are listed below:

• Adapter design pattern
  Adapter patterns create a connection between interfaces of different classes; let’s take the real-life example; we have two applications created with two different programming languages. Now, you want to compile both applications on a single platform. The adapter works here as it changes the interface of one instance so that other instances can easily understand it. The adapter hides all the complex conversions.
• Bridge design pattern
  The bridge design pattern splits the abstraction from its implementation. Let’s take the example of the main socket, which controls the lights, fan, TV, and AC of the house. This main socket acts as a bridge between the house’s sub-switches that control turning ON/OFF of the utilities.
• Composite design pattern
  Composite design patterns are used where a group of identical instances is considered as one instance. This pattern creates a hierarchal structure of similar objects. For example, the composite pattern creates a class that includes the faction of its objects.
  Let’s understand the pattern with an example of grouping the teachers into different departments, classes will contain all the details of a teacher like a name, salary, address, and department, and composite teacher class will use teacher class to add teachers in a different department and print the teachers.
• Decorator design pattern
  Decorator design patterns are used whenever new functions are needed to be added to existing objects without changing the entire structure of the class. Decorator design pattern creates a new class that abstracts the original class and provides the newly added functionalities. Let’s discuss an example of a decorator design pattern for a clearer understanding. Suppose we need to develop an online ordering system of a well-known burger restaurant that takes all requirements from a customer like what kind of bun, patty, and cheese they need in the burger and according to the requirement estimate the burger price.
  Since customer customization may vary from customer to customer, creating classes for each type of customization will be a difficult task. Here we can create a decorator pattern class, which will contain all the information to customize the burger, like adding extra cheese, extra patty etc. The main class will be the burger with the regular base price. By using a decorator class, the burger class can be extended and the price can be added according to the customization.
• Façade design pattern
  Façade design patterns act like abstraction, which conceals all the inner complex details of the application and provides a simple interface to the client to access the applications. Let us consider an example of a grocery shop, which is equipped with loads of items. When a customer visits the shop for the first time, he is unable to search and find the items he wishes to purchase. So he gives a list to the shopkeeper and the shopkeeper gives the customer all the required items. This is an example of Facade design pattern.
• Flyweight design pattern
  Flyweight design patterns, as the name implies, are patterns that fly away from the weight, i.e., a memory from the application. It reduces the objects from the source code to release the memory and boosts the performance. Flyweight design can be called code optimization. Flyweight searches the objects that are similar and use them. In case no match is found, it creates a new object. Improving system performance is a non-functional and essential requirement of software as it affects its appearance. Suppose we have an application running too slow due to lines and lines of code written and wastage of cache; now, flyweight design patterns work here. It will eliminate all the similar objects that are created and waste memory.
• Private class
  This type of pattern is used for security purposes where data are crucial. Create a private class and store the data, which will be in the same state throughout the code. All variables defined in the private class are only accessible within its class. To access the private class attributes, you need to create their getters and setters. The private class is used where information needs to be hidden from the outside world and only displayed to you.
• Proxy
  Proxy is all about providing access. In proxy design patterns, a class is created, which represents the functionality of other classes. Let’s take the example of a credit card that acts as a proxy and contains all your bank account amounts.

1.3 Machine Learning and Statistics

ML can forecast better and more accurate output. It builds the algorithms that use the input data statistics to predict the output. All social media websites use ML to display the data on the feed. The procedure of ML involves searching through data to look for patterns and program them accordingly. Some ML examples include ads displayed as a suggestion and fraud detection. Some of the ML methods are as follows:

1. Supervised machine learning algorithm
2. Supervised learning deals with the known and categorizes data. It can classify uncategorized data also. In supervised learning, a sample set contains input data with desired output data. Based on this new test, data can easily be categorized. For example, an application identifies the animal that is either an herbivore, a carnivore, or an omnivore animal. Using supervised learning, it already knew the classification of the animal. Now, whenever a new animal is entered into the system as an input, the system will automatically predict its category.
3. Unsupervised machine learning algorithm
4. In the unsupervised learning sample set, data are unknown and unlabeled. The data cannot be implemented directly as we are unaware of the outputs. It simply works on finding the similarities between the data and categorizes them as one. For example, categorize the customer based on which product they purchase, and based on a similar product, we can categorize customers.
5. Semi-supervised machine learning algorithm
6. Semi-supervised learning is the mixture of supervised and unsupervised learning as its dataset contains categorized and uncategorized data. The aim is to predict the new data that are more effective and accurate than the output data given by the user. For example, you wish to buy a product and watch ads related to the product, and suddenly, you want to review the same product. Though the product is from a different company, the categorized data would be the basic ad released by the company.
7. Reinforcement machine learning algorithm
8. In reinforcement learning, the machine does not learn from classified data; instead it learns from its experience and prediction, an agent takes all the actions (robotic avatar), finds all possible scenarios and fits in the best solution. Games such as hangman is the best example of reinforcement learning algorithm. Statistics is a representation of data, hypothesis, etc. It’s the correlation between the points, invariable and multivariable.
   Statistics types are as follows:
9. Forecasting continuous variable: If a variable can adapt value between minimum and maximum, then it is called continuous variable, for example, weight and age. It may be deterministic or probabilistic. A deterministic result can be compared and evaluated with respective observations. Probabilistic results are in the form of distribution.
10. Regression: It deals with the relationship between predictor and outcome variables. It analyzes the results based on predictive analysis and modeling. Let’s take the example of dieting apps that predict the diet schedule based on your daily food routine, weight, and height.
11. Classification: It’s the process of grouping the data on the basis of similar categories. For example, classify animals based on their habitats.

ML and statistics are two different methodologies, but how are they similar? The similarity between both of them is the prediction from the data. The primary difference between ML and statistics is manipulating the data; ML is simply working out the data using algorithms, whereas statistics is just mathematics in finding the patterns from the data.

1.4 Relat...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Acknowledgment
• About the Authors
• Introduction
• Chapter 1 Introduction
• Chapter 2 Overview of Python for Machine Learning
• Chapter 3 Data Analytics Life Cycle for Machine Learning
• Chapter 4 Unsupervised Learning
• Chapter 5 Supervised Learning: Regression
• Chapter 6 Supervised Learning: Classification
• Chapter 7 Feature Engineering
• Chapter 8 Reinforcement Engineering
• Chapter 9 Case Studies for Decision Sciences Using Python
• Appendix: Python Cheat Sheet for Machine Learning
• Bibliography
• Index