Machine Learning and Deep Learning for Applications: A Hands-On Study With Python

Introduction

In today's world, most people hear about machine learning, and they would imagine that this is a super robot or terminator that will destroy the world, but what really machine learning is? Arthur Samuel, a computer scientist who established the study of artificial intelligence, defined machine learning as “the study that gives computers the ability to learn without being explicitly programmed.” Machine learning is the science of computers, so they can learn from past experiences (input data) and make future predictions. Why do we pay much attention to machine learning? Because it implements intelligent alternatives to analyzing vast volumes of data. Machine learning can generate accurate results and analysis by developing fast and efficient algorithms and data-driven models for real-time data processing. Machine learning systems have different types and those types are useful to distinguish them in comprehensive sections, based on the following criteria:

• •

  Whether or not they are trained by human which categorize as supervised, unsupervised, semi-supervised, and reinforcement learning.

• •

  Whether or not they learn increasingly by themselves. (online or batch learning).

• •

  Whether they operate by simply analyzing new data points to known data points, or alternatively by identifying patterns in the training data and developing a predictive model and those are instance-based and model-based learning.

What does deep learning differ from traditional machine learning? Generally, deep learning is a subset of machine learning where algorithms are designed to function almost exactly as machine learning, but there are differences and deep learning has multiple levels (or layers) of artificial neural networks, each of them presenting a different understanding of the data it conveys or different functions. In simple words, it resembles the neural relation that exists in the human brain. For example, we have a collection of photos of apples and oranges. Assume we need to identify photos of apples and oranges separately using machine learning algorithms and deep learning models. From the machine learning point of view, the machine learning algorithm needs to present these apples and oranges images first as feature vector, and then classify these images according to the feature vector. But deep learning uses multiple layers to progressively extract higher-level features from the raw image input without representing each image by the feature vector.

Why does such a big difference exist between deep learning and traditional machine learning? The answer is the availability of structured data and different design of traditional machine learning algorithms and multiple-layered structure in deep learning. We need to mark the photos of apples and oranges to determine the components of both fruits. These data will be adequate for training a machine learning algorithm, and then the algorithm, later, continues to work on the basis that it understands the markings of many other photos of fruits, both on the grounds that the machine learning algorithm had studied earlier. From a deep learning point of view, the main difference is that deep learning doesn’t necessarily require structured/tagged image data (or feature vector) to classify two fruits. For solving this problem, the image data (apples and oranges) is fed through different levels of neural networks, and each network hierarchically defines the specific features of the images. After processing the data into different levels of neural networks, the system finds relevant identifiers to distinguish both fruits by their pictures.

“The best way to learn a thing was to do it”. In this chapter, we will learn machine learning and deep learning with application by a hands-on study in Python. We will have the data from UC Irvine Machine Learning Repository, Kaggle datasets, Amazon’s AWS datasets. Kaggle (Kaggle, 2010) is a repository for sharing ideas, competing, learning new information and coding tricks, and discussing various examples of real-world Machine Learning applications. In this chapter, our goal is to offer some problems and how we can solve them using Kaggle's available dataset. We will discuss Machine Learning categories, Supervised Learning, Unsupervised Learning, and Reinforcement Learning and will discuss some applications by step-by-step guidelines along with the methods. The project applications are listed below.