Real-World Applications of Machine Learning Algorithms

Name: Real-World Applications of Machine Learning Algorithms
Author: Briana Brownell

Discover the exciting ways machine learning algorithms can be used in everyday life

Preview the first lesson free — get full access to all 3 lessons.

Course: On-Demand

Thanks to our use of technology, most of us are exposed to machine learning whether we realize it or not. For example, let's consider spam emails. Since they're so commonplace, most email providers attempt to separate unsolicited and junk emails from users' regular email correspondence. But how does this work? There are several variables we might think about using to decide whether email is spam or not. First, consider the sender. Has this contact corresponded with the recipient before? Although past correspondence could be an indicator, we need more than just this information to decide whether the email is spam. Having had previous correspondence unfortunately doesn't necessarily mean all future correspondence will be spam free. The subject line and body of the email are also used to determine whether the email is spam or legitimate. We need to find a way to work with language data and turn it into something that can be used for prediction. Once we do that, we can use supervised machine learning to predict whether an email is spam. Think about how we might turn text in the columns. Text is made up of individual characters. None of the characters in English are good predictors of whether an email is spam because the meaning of the characters depends on words or phrases. Although this is true for English, it doesn't generalize to other languages. So remember that spam detection depends heavily on the language you're working in. With other indicators, like the length of the email in characters or words, we might not be so sure whether it would make a good predictive variable. This example illustrates how important it is for a data scientist to have at least some understanding of the domain of the data they're modeling. Turning text into data columns is usually done in a multi-step process. First, we need to separate the text into discrete chunks in a process called tokenization. In English, usually a token is a single word, but there are complications like contractions, hyphenated phrases, and spelling mistakes. To make a good predictive model, we need to figure out how to deal with these challenges. Each token becomes a column, and the presence or absence of a word becomes a factor to predict whether the email is spam. Some predictive models also use what are called in-grams or a series of words. This large matrix can then be used to predict whether an email is spam. After tokenization, we might consider processing those tokens to cut down the total number of words. For example, we may want to combine forms of the same verb together, like run and ran, or we might want to combine the singular with the plural of the same word like cat and cats. The trade-off using stimming or limitizing is that you combine the predictive power of all tokens that stim or limitize the same way together. That could make the variable less accurate as a predictor, but it also requires less data to train. In English, there are words that add very little to the meeting of context. We call those stop words. These are words like the and is, a, and are. Those words are unlikely to have much predictive power, but are extremely common in any text. There are several suggested stop word lists, but you may not want to use all the words on the list, or you might want to add additional words. Of course, the order of the words matters. Most newer language models also consider the order in which they appear. However, this adds a lot of complexity to the training and requires a lot more data than predicting just based on the words and email contains. To determine whether an email is spam, you also need to figure out an outcome variable. The practitioner needs to either label a data set of their own or use one of the many spam detection data sets available from various research teams. Working with language can be complicated in terms of pre-processing, but at its core, many of the core skills in building any machine learning model apply to these kind of cases.

Intermediate Provider : Briana Brownell 3 Lessons · 11m in English

Course Description

Supervised machine learning can be used to model language and predict whether or not an email is spam. Modelling language is a particularly challenging domain. In these lessons, we’ll discuss how to convert language into usable data using particularities of language. We’ll also explore why every language is different and requires a unique set of rules.

Businesses of all types can use customer segmentation to better understand their market needs and to build new products that address the unique behaviors and needs of customer groups. In these lessons, we will discuss the use of unsupervised learning to create a customer segmentation. We’ll also cover practical considerations of customer segmentations.

Finally, we will review neural-network based image classification. Because of the proliferation of image data, machine learning methods to classify images have gained popularity. Convolutional neural networks, a specific type of neural network created to address some of the difficulties in working with images, can provide an effective method to solve some of the most challenging problems in modern machine learning.

What You'll Learn

Convert language into usable data and apply supervised learning to detect whether an email is spam
Build customer segmentations using unsupervised learning to understand market needs and customer groups
Classify images using neural networks, including convolutional neural networks built to address image-data challenges
Identify the difficulties in working with different types of data
Explain the advantages of specific pre-processing and modelling methods
Define the business case for various styles of machine learning

Key Takeaways

Supervised machine learning can be used to model language and predict whether an email is spam.
Modelling language is a challenging domain because every language is different and requires a unique set of rules.
Businesses of all types can use unsupervised learning to create customer segmentations that reveal market needs and inform new products addressing the behaviors and needs of customer groups.
Convolutional neural networks are a specific type of neural network created to address some of the difficulties in working with images.
The proliferation of image data has increased the popularity of machine learning methods to classify images.

Frequently Asked Questions

What topics does this course cover?

The course covers three real-world applications of machine learning across three lessons: Spam Detection, Customer Segmentation, and Image Classification. It addresses converting language into usable data for supervised spam detection, using unsupervised learning to create customer segmentations, and neural-network based image classification including convolutional neural networks.

What will I be able to do after taking this course?

You will be able to identify the difficulties in working with different types of data, explain the advantages of specific pre-processing and modelling methods, and define the business case for various styles of machine learning.

Who can benefit from the customer segmentation portion of this course?

Businesses of all types can benefit, as the course discusses how customer segmentation helps them better understand their market needs and build new products that address the unique behaviors and needs of customer groups.

What skills does this course focus on?

The course focuses on applications of artificial intelligence, automated machine learning, machine learning, machine learning algorithms, machine learning methods, and supervised learning.

Why is modelling language described as challenging in this course?

The course explains that modelling language is a particularly challenging domain because every language is different and requires a unique set of rules when converting language into usable data.