The fundamentals of machine learning

ML creates models that can generalize new, previously unseen data. These models perform tasks such as classification, regression, clustering, and pattern recognition. Machine learning has numerous applications, including image and speech recognition, natural language processing, recommendation systems, medical diagnosis, finance, and autonomous vehicles.

Machine learning is also frequently associated with other AI subfields.

Deep learning vs. machine learning

Deep learning and machine learning allow computers to learn from data. But what is the difference between the two? And what is deep learning?

Deep learning is a subset of ML. Artificial neural networks with multiple layers extract high-level features from raw input data. These neural networks are modeled after the structure and function of the human brain, with interconnected layers of artificial neurons that process data.

The main differences between machine learning and deep learning are their architecture and capabilities:

• Structure: Deep learning extracts high-level features from raw input data into network layers, while machine learning algorithms parse data, learn from it, and make decisions or predictions that can be used to manually engineer.
• Learning: Deep learning algorithms automatically learn hierarchical representations of data, allowing them to discover intricate patterns and relationships directly from raw input. On the other hand, ML algorithms use either supervised learning, which trains a model on known input and output data, and unsupervised learning, which finds hidden patterns in input data.
• Complexity and scalability: ML algorithms may struggle to scale to such large datasets or capture complex relationships without extensive feature engineering. Deep learning models are capable of handling large amounts of data, making them ideal for tasks that require massive datasets and intricate patterns, such as image recognition, natural language processing, and speech recognition.
• Training and computation: Deep learning models typically require significant computational resources to train millions (or even billions) of parameters. By contrast, machine learning often has simpler training procedures and may require fewer computational resources.

Natural language processing

Natural language processing, or NLP, is a field of artificial intelligence that focuses on enabling computers to understand, interpret, and generate human language.

NLP tasks include:

• Text understanding
• Speech recognition
• Machine translation
• Text generation
• Question answering

Neural networks

Neural networks are a type of machine learning model based on the structure and function of the human brain. They are made up of interconnected nodes, known as neurons or units, which are organized into layers. Each neuron receives input signals, processes them with an activation function, and generates an output signal that is sent to other neurons in the network. This enables neural networks to perform functions like classification, regression, clustering, and pattern recognition.

How does machine learning work?

Machine learning is a subset of artificial intelligence that enables computer systems to learn and improve from experience without being explicitly programmed. The process involves feeding vast amounts of data into models and creating algorithms that allow them to recognize patterns, make decisions, and continuously improve their performance.

Machine learning relies on three fundamental components:

• Data: High-quality data is the foundation of machine learning. It can be structured or unstructured, and the more diverse and extensive the dataset, the better the algorithm can learn.
• Algorithms: Algorithms are rules and statistical models that analyze data, identify patterns, and make predictions. There are various types of machine-learning algorithms, each serving specific purposes.
• Computing power: Machine learning requires significant computing power to process data efficiently that is either provided by powerful hardware, cloud infrastructure, or both.

ML uses algorithms to analyze large datasets and identify patterns or relationships within them. The process typically includes the following steps:

• Data collection: The first step in machine learning is gathering relevant data from several sources. This data may consist of structured data from databases, unstructured data from text documents or images, or streaming data from sensors or IoT devices.
• Data preprocessing: Once collected, the data must be cleaned and prepared for analysis. This can include removing outliers, handling missing values, normalizing or scaling the data, and encoding categorical variables.
• Feature engineering: During this step, the variables that are most important to the problem are created from raw data. This can include transforming or combining existing features to create new ones that better reflect the data's underlying patterns.
• Algorithm selection: There are several types of machine learning algorithms, each of which is appropriate for a different set of tasks and data. Machine learning algorithms are classified into three types: supervised learning, unsupervised learning, and reinforcement learning.
• Model training: After selection, the algorithm is trained on the prepared dataset. During training, the algorithm learns from the data by adjusting its internal parameters to reduce the difference between its predicted and actual outcomes in the training data.
• Model evaluation: After training, the performance of the model is evaluated using a separate validation dataset or through cross-validation techniques. This helps determine how well the model generalizes new, previously unseen data and whether it effectively captures the data's underlying patterns.
• Model deployment: If the model performs well during evaluation, it can be used to make predictions or decisions based on new, real-world information. Deployment may include integrating the model into existing systems or applications, monitoring performance and reliability, and updating the model regularly as new data becomes available.

Data science vs. machine learning: Exploring key differences

Data science and machine learning are closely related fields, but they have distinct differences:

• Scope:
  • Data science is a broad term that refers to a variety of techniques and methodologies used to extract insights and knowledge from data. It includes data cleaning, visualization, statistical analysis, and predictive modeling.
  • ML is a subfield of data science that focuses on creating algorithms and models that can learn from data and make predictions or decisions.
• Focus:
  • Data science covers the entire data lifecycle, from data collection and preprocessing to analysis and interpretation. It involves understanding the data and applying various techniques to extract actionable insights.
  • Machine learning, on the other hand, focuses on developing predictive models using algorithms that can learn from data and improve over time.
• Techniques:
  • Data science analyzes and interprets data using a variety of statistical, mathematical, and computer science techniques. These methods may include regression analysis, clustering, classification, and natural language processing, among others.
  • On the other hand, machine learning focuses on algorithms and techniques that allow computers to learn from data and make predictions or decisions without being explicitly programmed.

Understanding the different types of machine learning models

ML focuses on three main types: supervised machine learning, unsupervised machine learning, and semi-supervised machine learning.

Although all these approaches enable computers to learn from data and make predictions, they can be used for different purposes.

Supervised machine learning

Supervised machine learning provides the algorithm with labeled data, meaning each data point is associated with a known output. The algorithm learns by comparing its predictions with the correct outputs and adjusting its parameters to minimize errors.

Supervised learning can be further categorized into two main types:

• Classification: The algorithm learns to divide input data into discrete categories or classes. For example, given an image of a handwritten digit, the algorithm may identify it as one of the digits 0–9.
• Regression: In regression tasks, the algorithm learns to predict continuous output values using input features. For example, by providing information about the size of a house, the algorithm may predict its price.

Unsupervised machine learning

In unsupervised machine learning, the algorithm must find patterns and relationships in unlabeled data independently. Clustering and dimensionality reduction are common applications of unsupervised learning.

This type of ML can be divided into three categories:

• Clustering: Clustering algorithms group similar data points together based on features to identify natural groupings or clusters within the data.
• Dimensionality reduction: Dimensionality reduction techniques seek to reduce the number of features in a dataset while retaining its key characteristics. These techniques help to simplify the data by removing noise or redundant information.
• Association rule learning: Association rule learning identifies relationships or associations among variables in a dataset. It identifies patterns like "if X, then Y," which can be useful for tasks like market basket analysis or recommendation systems.

Semi-supervised learning

Semi-supervised learning combines elements of supervised and unsupervised learning. It involves training the algorithm on a limited amount of labeled data and a more extensive amount of unlabeled data. This approach is useful when obtaining labeled data is expensive or time-consuming.

Reinforcement learning

In reinforcement learning an agent learns to make decisions based on interactions with an environment to achieve a specific goal. In this type of ML, the agent learns through trial and error, receiving feedback in the form of rewards or penalties based on its actions.

The process usually includes the following components:

• Agent: The entity or system that takes actions in the environment.
• Environment: The external system with which the agent interacts. It can be a physical or simulated environment.
• State: The current situation or configuration of the environment.
• Action: The decision or choice made by the agent at each time step.
• Reward: The feedback signal received by the agent after taking an action. It indicates how good or bad the action was in achieving the agent's goal.
• Policy: The strategy or set of rules that the agent uses to select actions based on the current state.

This approach can be applied to several domains, including robotics, autonomous vehicles, finance, healthcare, and recommendation systems.

Key differences between supervised and unsupervised machine learning

The primary difference between supervised and unsupervised learning is that supervised learning requires labeled data for training, while unsupervised learning does not. Supervised learning is used for tasks with clearly defined outputs, while unsupervised learning is suitable for exploring unknown patterns in data.

Here are the major differences:

1. Supervised learning:
   • The algorithm is trained on a labeled dataset, where each example is paired with a corresponding label or target variable.
   • The objective is to map input features to output labels or predictions based on the training data.
   • Supervised learning tasks include classification, where the model predicts a categorical label, and regression, where the model predicts a continuous value.
   • Examples of supervised learning algorithms include decision trees, support vector machines (SVM), linear regression, and neural networks.
   • During training, the model adjusts its parameters to minimize the difference between its predictions and the true labels in the training data.
2. Unsupervised learning:
   • In unsupervised learning, the algorithm is given an unlabeled dataset and must find patterns, structures, or relationships in the data without explicit guidance.
   • The objective is typically to discover hidden patterns or groupings in the data, such as clusters of similar data points or latent factors underlying the data distribution.
   • Unsupervised learning tasks include clustering, where the algorithm groups similar data points together, and dimensionality reduction, where the algorithm reduces the number of features while preserving the essential structure of the data.
   • Examples of unsupervised learning algorithms include k-means clustering, hierarchical clustering, principal component analysis (PCA), and autoencoders.
   • Unlike supervised learning, there are no explicit target variables to predict, and the model learns solely from the inherent structure of the data.

Common machine learning algorithms

Algorithms are the backbone of the entire machine learning process. Here are some of the most common ones:

• Linear regression: Used for predicting a continuous variable based on one or more input features. It fits a linear relationship between the input variables and the target variable.
• Logistics regression: Similar to linear regression, but used for classification tasks where the target variable is categorical. It models the probability that an instance belongs to a particular class.
• Decision tree: A tree-like structure where each internal node represents a feature, each branch represents a decision rule, and each leaf node represents a class label or a value. They are widely used for both classification and regression tasks.
• Support vector machine (SVM): A supervised learning algorithm that separates data points into different classes by finding the hyperplane that maximizes the margin between classes.
• Naive Bayes: A probabilistic classifier based on Bayes' theorem with the assumption of independence between features. It is commonly used for text classification tasks such as spam detection and sentiment analysis.
• K-nearest neighbors (KNN): A simple algorithm that stores all available cases and classifies new cases based on a similarity measure (for example., distance functions).
• Clustering: Used to group similar data points based on certain features or characteristics.
• Random forest: An ensemble learning method that builds multiple decision trees during training and outputs the mode or means prediction of the individual trees for classification or regression, respectively.
• Apriori: Used for association rule mining in transactional databases. It identifies frequent item sets and generates association rules based on the frequency of co-occurrence of items in transactions.
• Principal component analysis (PCA): A dimensionality reduction technique used to reduce the number of variables in a dataset while preserving the most important information.

ML algorithms are used in various applications such as classification, regression, clustering, dimensionality reduction, recommendation systems, and anomaly detection. The choice of algorithm depends on the nature of the problem, the characteristics of the data, and the specific requirements of the application.

Machine learning applications and examples

Machine learning has numerous applications across various industries. Some common examples include:

• Recommendation systems: Websites such as Amazon, Netflix, and Spotify use ML algorithms to analyze user preferences and make personalized recommendations for products, movies, and music.
• Predictive maintenance: Manufacturing companies use machine learning to predict equipment failures and schedule maintenance ahead of time, reducing downtime and costs.
• Fraud detection: Banks and financial institutions use machine learning to detect fraudulent transactions by analyzing patterns and anomalies in transaction data.
• Image recognition: Companies in multiple industries that rely on facial recognition, object detection, and medical image analysis use machine learning algorithms to accurately interpret and classify images.
• Virtual assistants: Assistants like Siri and chatbots use NLP algorithms to understand and respond to human language, enabling natural and intuitive interactions.
• Healthcare diagnostics: In healthcare, ML is used to diagnose diseases based on medical images, predict patient outcomes, and provide personalized treatment recommendations.
• Autonomous vehicles: Self-driving cars use machine learning algorithms to perceive their surroundings, navigate, and make real-time decisions using sensor data.
• Customer churn prediction: Businesses use ML to predict customer churn by analyzing historical data and identifying factors that influence customer attrition.

Machine learning with OutSystems

Organizations can unlock the transformative power of machine learning with OutSystems. The OutSystems low-code platform is powered by powerful AI services that automate, guide, and validate development. AI and ML enable development pros to be more productive and guide beginners as they learn, all while ensuring that high-quality applications are delivered fast and with confidence. By embedding the expertise and ML gleaned from analyzing millions of patterns into the platform, OutSystems has opened up the field of application development to more people.

Understanding the different types and algorithms of machine learning is essential to unlocking its full potential in your applications. OutSystems makes that easier by providing connectors to machine learning services that revolutionize how your customers interact with technology and make decisions. As a result, the future of low-code application development is even more promising, offering endless possibilities to create intelligent and transformative solutions. Embrace the power of machine learning and stay ahead in the digital era with OutSystems.

Machine learning frequently asked questions