30.6.1. Popular Algorithms

• Regression Algorithms: Linear, Logistic

Regression algorithms are used to predict continuous outcomes. Linear regression aims to find the best-fit line through data points, essentially modelling the relationship between a dependent variable and one or more independent variables. Logistic regression is used when the outcome is binary, meaning it predicts categorical results (like yes/no or 0/1). This is useful in scenarios such as classifying whether a construction component will pass or fail a stress test based on various input factors.

Imagine you're a baker trying to figure out how many cookies to bake based on how many customers usually come in on a rainy day versus a sunny day. Linear regression helps you make predictions based on past data, while logistic regression would help you understand whether to expect a busy or quiet day based on the weather.

• Classification Algorithms: Decision Trees, k-NN, Naive Bayes

Classification algorithms are used in scenarios where you want to categorize data into distinct classes. Decision Trees involve creating a model that predicts the value of a target variable based on several input variables. The model splits data into branches to make decisions. The k-Nearest Neighbors (k-NN) algorithm classifies data points based on how closely they resemble other data points, with 'k' indicating the number of neighbors considered. Naive Bayes is based on applying Bayes' theorem with the assumption that all predictors are independent, effectively simplifying calculations. These algorithms are useful in tasks such as determining the type of construction material needed based on various properties.

Think of a tree diagram that helps you decide what type of fruit to eat based on its color and size. As you encounter different fruits, you might use k-NN by looking at the most similar fruits you've seen before to make your choice. For Naive Bayes, consider sorting emails into 'spam' or 'not spam'; the algorithm quickly assesses features of past emails to categorize incoming ones.

• Clustering Algorithms: K-Means, DBSCAN

Clustering algorithms are employed when you want to group similar data points into clusters. The K-Means algorithm partitions data into 'k' predefined distinct non-overlapping subgroups (clusters), thereby minimizing the variance within each cluster. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clusters data based on the density of data points, effectively identifying clusters of varying shapes and sizes while ignoring noise. These methods can be applied in civil engineering for detecting patterns in soil samples or clustering types of construction materials based on their properties.

Imagine a teacher grouping students based on shared interests or performance levels. K-Means is like deciding beforehand how many groups to form, while DBSCAN allows for flexible grouping based on how many students gather around common interests, potentially ignoring those students who simply don’t fit into social cliques.

• Neural Networks: CNNs for image recognition, RNNs for time-series forecasting

Neural networks are computing systems inspired by the human brain and are particularly effective in recognizing patterns in data. Convolutional Neural Networks (CNNs) excel in processing structured grid data like images, making them ideal for tasks such as identifying defects in construction images. Recurrent Neural Networks (RNNs) are designed for sequential data, making them suitable for time-series forecasting, like predicting material wear and tear over time. Their architecture allows them to retain information from prior inputs, enabling complex tasks.

Consider how we recognize a face. CNNs work similarly, breaking down an image into features, such as the shape of the eyes or the curve of the mouth, which helps us identify the person. RNNs work like remembering a melody over time, where each note influences the next one, allowing you to keep track of the entire song even as you hear one note at a time.

--