Introduction to Machine Learning

In this lesson, you'll learn about the three main types of machine learning analytics—descriptive, predictive, and prescriptive—and how they enable ML to drive disruption in many industries. You'll also explore the kind of problems that machine learning can help solve and the key considerations when selecting data for a machine learning project.

The machine learning pipeline, from data pre-processing to feature engineering and model selection, centers on data. You'll find out how data is selected and cleaned up for use, and how data scientists decide which features to include. You'll also learn how they go about creating the algorithms that will yield accurate output.

This lesson focuses more closely on the data that feeds the machine learning process. Data scientists spend up to 80% of their time in data-preparation related tasks. You'll learn about the main techniques used for data preparation purposes, including cleaning, encoding, scaling, and correcting imbalances, to get the most relevant and error-free data to train a machine learning model.

Supervised learning is one type of machine learning that maps labeled input data to known output. By finding the relationship between the input and the output, the system can apply that relationship to other input to predict the output. This lesson takes a quick look at the mathematics behind how the system finds that relationship using linear, polynomial, or logistic regression.

Regression enables a system to find the relationship between numeric inputs and outputs. But when the data is not numeric, a classification algorithm works to predict the category that data belongs to. Classification is an important task since it allows the computer to choose among different alternatives. In this lesson, you'll learn about binary, multi-class, and multi-label classification.

Ensemble methods of machine learning combine several simple models with weak predicting power in order to get better predictions. Akin to the idea that two heads are better than one, these methods aggregate the results of many predictions. We'll look at a range of ensemble methods, including voting, averaging, weighted averaging, bagging and bootstrap aggregating, random forest, and adaptive boosting, along with some practical examples of how they are used.

Unsupervised learning is a type of machine learning that deals with unlabeled datasets; it finds structure in data without having information about the correct output. In other words, unsupervised learning seeks to describe data as opposed to predict data (as is the case with supervised learning). In this lesson, you will learn about clustering algorithms and dimensionality reduction, two techniques for unsupervised learning, along with some application examples.

Semi-supervised learning is a machine learning method that combines the best of supervised and unsupervised learning in terms of both data availability and outcomes. It uses both labeled and unlabeled data and actually closely mimics how humans learn. It can even be trained to label data that is used to train other algorithms. This lesson will cover self-training, pseudo-labels, and transfer learning. It will also look at practical examples of how semi-supervised learning is used.

Reinforcement learning is a type of machine learning where the system learns through interacting with its environment, not by having access to large amounts of training data. In this lesson, you'll explore what it means for a computer to interact with the environment, how to model and formalize these interactions, and how machines learn in this context.

A successful ML learning project requires the project staff to work through a set of steps, collectively known as the machine learning workflow. In this lesson, you'll look at the final two steps in the process: training and deployment. We'll look at the difference between offline and online training and predictions, automated machine learning, and how the cloud environment affects machine learning functions. You'll also learn about model and data versioning, testing, and data validation, all of which are important to the deployment process.

Machine learning is a very active research area, and its impact on businesses and our daily lives has both increased and become more evident during the last decade. As the field further advances, developments in data management and computing capacity will play an important role. In this lesson, you'll explore some of the most prominent active areas in machine learning and which future improvements are likely to move the field forward.