Fun with Machine Learning: Simplify the Data Science process by automating repetitive and complex tasks using AutoML (English Edition)

By Arockia Liborious and Dr. Rik Das

“Fun with Machine Learning” is an essential guide for anyone looking to learn about machine learning and how it can be used to make informed business decisions.

The book covers the basics of machine learning, providing an overview of key concepts and terminology. To fully understand machine learning, it is important to have a basic understanding of statistics and mathematics. The book provides a simple introduction to these topics, making it easy for you to understand the core concepts. One of the key features of the book is its focus on AutoML tools. It introduces you to different AutoML tools and explains how to use them to simplify the data science processes. The book also shows how machine learning can be used to solve real-world business problems, such as predicting customer churn, detecting fraud, and optimizing marketing campaigns.

By the end of the book, you will be able to transform raw data into actionable insights with machine learning.

• Language: English
• Publisher: BPB Online LLP
• Release date: Mar 23, 2023
• ISBN: 9789355517845

Book preview

CHAPTER 1

Significance of Machine Learning in Today’s Business

Let us be purpose-driven and empowered with data and ethics.

Everyone in today’s fast-moving digital world wishes to be data-driven and wishes to create value. Do we really need to be data-driven, or can data even drive things? A million-dollar question, right? Yes indeed. Humans are emotional beings and always need a just cause or purpose, as often cited by leadership and management guru Simon Sinek. Data can unravel the mystery of whether we are progressing toward our cause but not where to go. As leaders, students, managers, and decision makers, it is imperative to use data to power our purpose. The issue of poor data analysis has plagued humanity for some time, but it has become increasingly apparent in the current era of widespread digital transformation and interconnectedness.. Everyone must know their way around data and be comfortable talking about it. This chapter will explain that the need for insights from data is stronger than ever before.

Structure

In this chapter, we will cover the following topics:

Hype behind machine learning and data science

Benefits of machine learning in business introducing data

Types of data in business context

Challenges with data

Citizen data science

Data science for leaders

Objectives

After studying this chapter, you should be able to relate how data plays a critical role in the business decisions we make. The chapter will also help you understand how machine learning is helping improve the decision-making process and how to utilize data to our advantage and make decisions based on data insights.

Hype behind machine learning and data science

Imagine the time you first learnt how to add two numbers in your Mathematics class in school. For some years, you would manually add up numbers till you had to use a calculator to perform the same task. Did it mean you could not do it manually any longer? No, it was because the process of calculation had to be quickened so that you could focus on the other critical steps of the problem.

In business too, we can take decisions based on experience and suggestions from experts we trust. However, is this the right step that would help us take faster and accurate decisions in the new normal world that is heavily data-driven? So, the question is, ‘Do you want to spend your valuable time in tasks that can be automated, or do you want to spend time utilizing the data insights to make critical decisions? If you want to do the latter, then you made the right choice by getting this book.

We will help you make the best use of your time and effort to utilize data, wrangle it fast and then derive relevant insights from the data. Now let us walk through the history of machine learning and look at how it has evolved over the years. Its origin can be traced back to the 17th century, when people were trying to make sense of data and process to make quick decisions. A simple evolution chart depicting the machine learning journey is shown in Figure 1.1:

Figure 1.1: History of Machine Learning

Blaise Pascal created the very first mechanical adding machine in 1642. Next, the data storage challenge was overcome using a weaving loom to store data by Joseph Marie Jacquard. Over time, we developed concepts like Boolean logic, statistical calculations, and the Turing test to evaluate whether a computer had intelligence, and eventually, the phrase artificial intelligence (AI). After a series of other inventions, recent years have seen advancements in machine learning algorithms. Today, three major innovations, as listed here, have fuelled the recent buzz and helped companies and individuals use and experiment machine learning technologies. In a nutshell, they have democratized machine learning to all:

Higher volume of data and cheap storage: Business-critical applications are producing and storing more data than ever before, thanks to cloud-based tools and the decreasing cost of storing data through services like Google Cloud Storage, Amazon Redshift, Microsoft Azure Services, and others. Most of these tools are highly intuitive and user friendly, with easy-to-use click and move features that simplify your work process incredibly.

Open-source: Open-source machine learning libraries, such as scikit-learn, Google’s TensorFlow and Orange, make cutting-edge algorithms more usable and accessible to a larger community of data scientists and engineers.

Greater computing power: With the advent of cloud-based technologies and custom hardware designed for machine learning, these systems can now run faster and at a lower cost, making them more suitable for a wide range of business needs.

Consider machine learning in this light. You, as a person and as a user of technology, carry out such actions, which allow you to make a decisive judgement and classify something. Machine learning has advanced to the point that it can mimic the pattern-matching ability of human brains. Algorithms are now used to teach machines how to recognise features of an object.

To provide just one example, a computer may be shown a cricket ball and instructed to treat it as such. The programme then uses the data to identify the different characteristics of a cricket ball, each time adding new data to the mix. Initially, a machine could identify a cricket ball as round and construct a model that states that everything round is a cricket ball. The programme then discovers that if anything is round and red, it is a cricket ball, when a red colour ball is added later. Then, a reddish-brown colour ball is introduced, and so on.

The machine must update its model as new knowledge becomes available and assign a predictive value to each model, indicating the degree of certainty that an entity is one item over another. Here, predictive value refers to the probability of identifying the ball correctly as cricket or tennis ball. As you can see in Figure 1.2, a machine learns the information provided to it by the user, executes certain action and receives feedback for it, and the learning continues as a feedback loop. The learning step is called as model training, and the feedback loop is called model retraining.

Figure 1.2: How does machine learning work?

The following figure gives an overview of the most common types of machine learning techniques available today: supervised learning, unsupervised learning, and reinforcement learning:

Figure 1.3: Types of Machine Learning

Supervised Learning

When an algorithm is trained to predict an output (also called a label or target) from one or more inputs (also called features or predictors), we say that the algorithm is engaging in supervised learning. As the name implies, supervised learning occurs when the algorithm is given labelled training instances that consist of input-output pairs. The algorithm’s objective is to generalise from the training examples and provide reliable predictions on novel, unseen data. For example, to design a face detection algorithm, you can feed or train the machine with images of people, animals, structures etc., along with their labels, to a point where the machine can accurately recognize a face in an unlabelled image. Figure 1.4 depicts this learning process where the machine is provided with an image of mangoes labelled as mango, learns various hidden features and then predicts in real time whether an image contains mango when an unlabelled image is given.

Figure 1.4: Supervised Learning

Unsupervised learning

Unlabelled data is analysed and categorised by the computer based on similarities found. Unlike the previous example, you provide the image but without labels. Still, the machine will be able to group the images based on certain common features (for example, texture, edges, and colour). The missing piece, however, is whether spherical objects qualify as faces.

The primary purpose of these algorithms is to unearth previously unseen patterns in data. In the following image, you can see how unlabelled data is provided, and the algorithm is left to discover patterns and relationships within the data on its own to cluster the data such that it makes sense.

Figure 1.5: Unsupervised Learning

Reinforcement learning

Reinforcement learning is a form of machine learning technique that is unique and totally different from supervised and unsupervised learning techniques. We use the principle of giving incentives for any good outcome as the foundation of our algorithm.

Let us use the example of a dog to make it clearer. We may teach our dog to perform specific acts, but it will not be easy. You would tell the dog to perform certain tasks, and with each successful completion, you would reward them with a biscuit. The dog will recall that if it performs a specific action, it will be rewarded with biscuits. This will ensure that it follows the instructions correctly the next time. Here, we impose or attempt to impose a correct action in a specific manner. In a nutshell, reinforcement learning is a form of learning technique in which we reward the algorithm with feedback so that it can learn from it and enhance future performance. In the following image, you can see how the machine can learn in a real environment by making mistakes and learning from it through a feedback loop, just like a human learning to ride a bicycle.

Figure 1.6: Reinforcement Learning

Benefits of machine learning in business

Many businesses dealing with large quantities of data have acknowledged the value of machine learning. Starting from the talent hiring process to achieving favorable business results, everything is hugely data-driven. For example, a candidate’s resume for a job role passes through an automated Applicant Tracking System that matches the keywords in the resume with the keywords in the job description to select or reject a candidate.

Organizations can operate more effectively or gain an advantage over rivals and competition by carefully extracting information from this data – often in real time. Can you picture how many people who want to purchase Nike shoes online end up on Amazon instead? Amazon, like most other online retailers, leverages customer search data and keywords from other sites to fine-tune its own website’s keyword strategy and attract more shoppers.

Try this: Search Buy Nike Shoes ‘online in your city, check which website and what offers show up for you. For each person, depending on their past searches for related objects, it will be different. This is the strength of data, which has the ability to transform businesses. Machine learning’s ability to scale through a wide range of fields or domains, such as contract management, customer service, finance, legal, distribution, quote-to-cash, quality and pricing, is due to its ability to learn and evolve over time.

Machine learning has several completely realistic applications that can lead to tangible business outcomes – such as saving time and money – that can have a significant impact on the company’s future and resource optimization plans. At critical interaction points, we are seeing a huge effect in the customer service industry, where machine learning is helping people get things done faster and more effectively.

Machine learning automates tasks, such as updating login credentials, which would otherwise entail support from a live agent. Every website, from Policy Bazar to Pepper fry, now has a chatbot. This frees up staff time to provide the type of individualized attention to customers that can’t be replicated by a machine.

The most powerful means of engaging trillions of users in social media is machine learning. Machine learning is at the core of all social media sites for their own and consumer benefits, from personalising news feeds to making targeted advertisements.

The most common application of machine learning that uses image recognition is Facebook’s and Apple iPhone’s auto-tagging function. This is something you might have noticed while saving or uploading photos if you are a regular user of Facebook or Apple. Facebook’s face recognition can identify your friend’s face with only a few manually tagged photos (almost equalling human capabilities). The three phases of image recognition are detection, classification, and identification. Detection is the process of analysing a picture to spot certain objects within the image. The terms Image Detection and Image Classification are sometimes used interchangeably. Classification is used to categorise objects within the image or the image itself, while detection can be used if your aim is to just count the number of objects within the image without knowing what the object is. Identification is the process of analysing the likeliness of a face or object match between two or more photos of the same individual or item. Other uses of facial recognition systems are attendance tracking, airport authorities to verify passenger information, and so on. However, one has to be mindful about the bias that goes into these applications as well. We would highly recommend watching the Netflix documentary Coded Bias to know more about this.

Combating fraud is one of the most popular applications of machine learning in the banking and finance industry. Machine learning is ideally suited for this use case because it can sift through massive volumes of transactional data and spot anomalies. Any transaction a customer makes is evaluated in real time and assigned a fraud score, which indicates the chances of the transaction being fraudulent. In the event of a fraud transaction, depending on the severity of fraud-like trends, the transaction is either blocked or handed over for manual inspection. When you swipe your card for a large transaction, you would get a call from your bank requesting for a confirmation on the transaction. This is a real-life example of how financial data is being used by the bank to identify a suspicious