What is machine learning?

Forging a career in information technology means potentially working with some of the most cutting-edge technology available today. Machine learning, a segment of artificial intelligence, employs algorithms to input data that's already been proven to predict new output values. As explained in our blog on artificial intelligence: With machine learning, software applications are programmed to become more accurate over time at predicting outcomes.¹

Machine learning is complex but can be better understood when broken down into its different models and applications.

Machine learning models

Machine learning isn't just one set process. It can be broken up into different models to be used depending on your intended application, what type of data you're using and so on. To begin understanding machine learning models, you should first understand the two types of machine learning approaches: supervised learning and unsupervised learning.

Supervised machine learning

According to IBM, supervised learning can be defined by its use of labeled datasets.2 By using these established data points, algorithms can be trained, or "supervised," into classifying data and predicting outcomes accurately. Because the algorithm can measure its accuracy by the labeled data points, it's able to become more accurate and learn over time.²

IBM gives the example that machine learning can be used to predict your travel time based on the time of day, weather and so on but first the algorithm needs the data points that train it to understand that traffic is heavier in the morning and evening or that snow can cause delays.²

Unsupervised learning

Unsupervised machine learning uses algorithms to analyze and cluster unlabeled data. Rather than using labeled training data to become more accurate, these machine learning programs uncover patterns without the need for training or "supervision." Although this type of machine learning model works without interference to discover the inherent structure of data, it does eventually require human assistance to validate outputs. ²

For example, an algorithm for an online retail shop can identify that shoppers often purchase groups of products at the same time. You would want a data scientist or other person, however, to validate the algorithm so that it shows beanies and mittens to someone who has put a winter coat in their cart instead of straw hats and sandals.

Deep learning

Deep learning is a type of machine learning that uses large amounts of data and several layers of algorithms to "simulate the behavior of the human brain" although for now, technology still remains a ways off from matching human ability. Deep learning is able to take in unstructured or unlabeled data and automate feature extraction. IBM gives this example to explain deep learning:³

Let’s say that we had a set of photos of different pets, and we wanted to categorize by “cat”, “dog”, “hamster”, et cetera. Deep learning algorithms can determine which features (e.g. ears) are most important to distinguish each animal from another.³

Common applications of machine learning systems

How is machine learning typically employed in our day-to-day life? There are dozens of ways machine learning technology is changing the way we interact with the world. Here are just a few examples:^4,5

• Image recognition: Have you ever been surprised by your mobile photo application recognizing the face of one of your friends? That's machine learning put to work. Machine learning is also capable of labeling x-rays as cancerous and recognizing handwriting. Yelp even uses this type of machine learning to identify user photos and sort them into categories like food, menus and restaurant photos.
• Speech recognition: Another common example of machine learning that's right on your phone: voice-to-text. Voice search, voice dialing and appliance control like is used with Google's Home device or Amazon Alexa are all machine learning examples.
• Medical diagnostics: Machine learning has been embedded into electronic health record systems to help formulate diagnoses and recognize patterns in symptoms. Along with image recognition that can identify cancerous cells, machine learning can also even help analyze bodily fluids.
• Detecting money laundering and fraud: Capital One uses machine learning to detect, diagnose and remediate spending behavior in real time to detect stolen credit cards and other forms of fraud.
• Timing user notifications: Duolingo uses machine learning to help its users learn new languages. The app analyzes answers to predict how long the user will remember a certain word. It then uses that data to determine when to suggest you do a refresher lesson.
• Recommendations: One of the ways you are most likely interacting with machine learning every day is through the filtering of apps to recommend content by analyzing viewing habits of users and comparing them to make informed suggestions.

Machine learning algorithms

To get more technical, let's dig into machine learning algorithms. Under the umbrellas of supervised learning and unsupervised learning are different types of algorithms that can accomplish tasks based on your goals and available data. Below are the basics on the most commonly used algorithms but you can learn much more about these algorithms and how to appropriately apply them in a machine learning course.

Supervised machine learning algorithms:⁶

Supervised learning algorithms can be split into two groups: linear models and tree-based models.

Linear models include:

• Linear regression: A simple algorithm that models a linear relationship between inputs and a continuous numerical output variable. Applications include prediction of stock prices, housing prices and a customer's lifetime value.
• Logistic regression: A simple algorithm that models a linear relationship between inputs and a categorical output (1 or 0). Applications include predicting credit risk score and customer churn prediction.
• Ride regression: Ride regression is a type of machine learning that penalizes features that have low predictive outcomes by shrinking their coefficients closer to zero. Can be used for classification or regression algorithms. Applications include predictive maintenance for automobiles and sales revenue predictions.
• Lasso regression: This type of machine learning penalizes features that have low predictive outcomes by shrinking their coefficients to zero. Can be used for classification or regression algorithms. Applications include predicting housing prices and clinical outcomes based on health data.

Tree-based models include:

• Decision tree: Decision Tree models make decision rules on the features to produce predictions. It can be used for classification or regression. Applications include customer churn prediction, credit score modeling and disease prediction.
• Random forests: This machine learning algorithm combines the output of multiple decision trees. Applications include credit score modeling and predicting housing prices.
• Gradient boosting regression: Gradient boosting regression employs boosting to make predictive models from an ensemble of weak predictive learners. Applications include predicting car emissions as well as ride-hailing fare prices.
• XGBoost: Gradient boosting algorithm that is efficient and flexible. Can be used for both classification and regression tasks. Applications include churn prediction and claims processing in insurance.
• LightGBM Regressor: A gradient boosting framework that is designed to be more efficient than other implementations. Applications include predicting flight time for airlines and predicting cholesterol levels based on health data.

Unsupervised machine learning algorithms:⁶

Unsupervised machine learning algorithms can also be broken into two groups: clustering and association.

Clustering models include:

• K-Means: The most widely used clustering approach—it determines K clusters based on euclidean distances. Applications include customer segmentation and recommendation systems.
• Hierarchical clustering: A "bottom-up" approach where each data point is treated as its own cluster—and then the closest two clusters are merged together iteratively. Applications include fraud detection and document clustering base don similarity.
• Gaussian mixture models: A probabilistic model for modeling normally distributed clusters within a dataset. Applications include customer segmentation and recommendation systems.

Association models include:

• Apriori algorithm: Rule based approach that identifies the most frequent itemset in a given dataset where prior knowledge of frequent itemset properties is used. Applications include product placements, recommendation engines and promotion optimization.

Become an expert in machine learning and its applications

If you're fascinated by the myriad uses of machine learning, consider earning an online Master of Science in Computer and Information Science from Marquette University. Marquette's faculty includes experts in machine learning and artificial intelligence who will help you learn the intricacies of machine learning as well as how it can be used to really make an impact on the world. Schedule a call with an Admissions Advisor to find out more about the online computing master's program and how it can help you to forge a career in machine learning.