Machine learning demystified: algorithms and applications

For some people, the idea of ‘machine learning’ evokes fear. It can be a worrying thought that someday computers and robots are clever enough to do our jobs, find out everything about us, or, at worst, might take over the world.

What machine learning makes possible however, both in what it technologically enables and the job opportunities it can create, are vast and exciting. 

In the last few years the fascinating world of machine learning has started to open up. We are already benefiting from it in many aspects of our lives, from healthcare technologies that support better disease diagnosis, to speech recognition software that provides wider access to digital tech.

What is machine learning?

Machine learning is an area of artificial intelligence that uses algorithms trained on data sets to create models. 

Machine learning models allow computers to perform tasks that would otherwise only be possible for humans, like categorising images, or translating speech into text.

In a nutshell, the computer learns from datasets and uses the information to perform new tasks without needing to be instructed by a computer program.

To learn, the computer uses computational models, or algorithms, to carry out data analysis, understand patterns and make predictions. Machine learning algorithms come in many shapes and sizes and vary depending on the requirements of the end user, from detecting spam messages to making personalised film recommendations, or predicting who might get a particular disease.

Different machine learning algorithms support different applications

Supervised Learning Algorithms

Supervised learning algorithms can be used to predict continuous values like house prices and assign labels to data points for image classification, or detecting spam emails.

Types of supervised learning algorithms using labeled data include linear regression, logistic regression, support vector machines (SVM), decision trees and random forests.

Unsupervised Learning Algorithms

These include algorithms that cluster similar data points together and are useful for customer segmentation and image classification. Other unsupervised learning algorithms reduce the number of features in unlabeled data sets while preserving information, a method known as dimensionality reduction.

Classification algorithms can also identify unusual patterns, which is useful in fraud detection.

Different algorithms in this group include clustering algorithms such as K-means clustering, and hierarchical clustering and principal component analysis (PCA).

Ensemble Learning Algorithms

This type of machine learning algorithm combines multiple models to improve performance. Examples include: Random Forest, AdaBoost and XGBoost.

Deep Learning Algorithms

Deep learning algorithms have multiple layers called neural networks that can be used for image recognition, natural language processing, and speech recognition.

Neural networks mimic how neurons in our brains signal to one another. They consist of interconnected nodes, organised into layers. A neural network usually has an input layer, several hidden layers and an output layer.

Examples of deep learning methods include deep neural networks and convolutional neural networks that simulate complex decision-making, similar to how our brain works.

Reinforcement Learning algorithms

Even more advanced and complex algorithms can train programmes to make sequential decisions based on rewards. This type of algorithm is used in game design and robotics.

Other Algorithms

• Naive Bayes – used for text classification and sentiment analysis.
• Time Series Forecasting – predicts future values based on historical data.
• Recommendation Systems – allows personalised content recommendations, for example, on Netflix and Amazon.

Different algorithms are often used together in machine learning frameworks to develop applications suited to specific tasks.

What are the challenges for optimisation of machine learning algorithms and applications?

Training data quality

High quality data is crucial to allow machine learning algorithms to be trained to make good predictions. Preprocessing of input data is often needed to remove outliers and handle missing values. If there is insufficient training data available, the model cannot work or develop.

Underfitting and overfitting training data

In the same way that sometimes our clothes can be too tight or too loose and make us uncomfortable or self-conscious, in data science, data that does not fit the size of the algorithm or computational model can mean the machine learning algorithm cannot function well.

Underfitting happens when the model is too simple to capture underlying patterns in the data. Whereas overfitting occurs when too much data leads to a negative impact on performance of the algorithm.

Building machine learning applications

Machine learning is a complex business and to build a machine learning application requires an understanding of different algorithms so the most suitable one is chosen.

Expertise is needed to optimise the parameters and develop code and parallel computations to speed up processes.

Monitoring is also important as algorithms can start to behave unexpectedly as datasets grow.

How are machine learning algorithms and applications being used today?

Machine learning is a rapidly progressing field and applications are already being seen in sectors ranging from healthcare to data security and weather forecasting.

As the size of datasets continue to increase and machine learning tools evolve, large data sets can be analysed in real-time for a range of metrics, and machine learning techniques streamlined to make ever better predictions and recommendations. 

The insights they provide can support better business decision-making, improve diagnostics in healthcare, provide enhanced accessibility to digital platforms for people with disabilities and facilitate accurate forecasting to help avoid natural disasters. The following use cases give a flavour of how machine learning is driving varied advancements across sectors:

Healthcare

Machine learning algorithms are increasingly being used on large datasets of medical images to predict which features indicate disease. Machine learning is also helping in the validation of new drug targets.

Retail

Recommender systems use machine learning to suggest products a consumer may like based on previous behaviour and preferences.

Security and fraud  

Machine learning models can authenticate images and help detect fraud in accounting systems.

Finance and banking  

Chatbots handle customer queries and trading systems increasingly rely on machine learning and predictive modelling.

Transport and autonomous vehicles

From route planning to driverless cars, machine learning is changing the way we travel.

Image and speech recognition

Virtual assistants like Alexa and Siri have become commonplace, understanding and responding to the spoken word.

Social media

Personalised content on social media platforms appears thanks to the training processes implemented by machine learning models that ensure we see and hear content according to our preferences.

Environmental conservation

Machine learning algorithms are being used to predict climate patterns, track wildlife and monitor deforestation.

What’s next for machine learning and artificial intelligence?

As machine learning and artificial intelligence tools advance, the field of computer science continues to expand to meet the demand for innovation.

The 100% online MSc Computer Science at City, University of London is designed to equip those set on developing real-world skills in artificial intelligence, and the principles of machine learning for the careers of the future.

This master’s degree provides highly in-demand skills, preparing you to work as a digital professional across many sectors. These include finance and banking, technology and start-ups, education, the civil service and more.