DATA 155 Survey to Data Science - Module 6: Machine Learning Techniques

Vocabulary practice flashcards covering supervised, unsupervised, and semi-supervised learning setups, classification metrics, algorithms, support vector machines, clustering methods, dimensionality reduction, and neural network fundamentals.

Generalization

Useful performance of a machine learning model on new, unseen data.

Supervised Learning

A learning setup where a model learns from labeled data (e.g., emails labeled spam or not spam) with the goal of predicting a label for new instances.

Unsupervised Learning

A learning setup where an algorithm works on data with no target labels (e.g., raw email text) to discover patterns, groups of similar topics, or anomalies.

Semi-Supervised Learning

A learning setup that leverages both a small set of labeled data and a large set of unlabeled data (e.g., $100$ labeled and $10{,}000$ unlabeled emails) to learn a classifier.

Regression

A task where features supply inputs to predict a continuous numeric target (e.g., predicting a house sale price of $\text{\$}240{,}000$).

Classification

A task where features supply inputs to predict a discrete categorical target label (e.g., price bands such as low, medium, or high).

Binary Classification

A classification task that predicts an outcome between exactly two classes (e.g., spam or not spam, benign or malignant).

Multiclass Classification

A classification problem where the categorical target has more than two possible classes.

Precision

An evaluation metric that measures the accuracy of positive flags, calculated as the ratio of true positive flags to total flags raised (e.g., $\frac{8}{10} = 80\%$).

Recall

An evaluation metric that measures how much of the target class is caught, calculated as the ratio of true positives caught to total actual positives (e.g., $\frac{8}{20} = 40\%$).

Decision Tree

A classification method that follows a sequence of learned feature questions down branches to arrive at a predicted label.

K-Nearest Neighbors (KNN)

A classification method that classifies a new example by evaluating nearby labeled examples and taking a majority vote among the $K$ nearest neighbors.

Logistic Regression

A classification algorithm that estimates the probability of class membership and applies a decision threshold (e.g., $0.50$) to make a prediction.

Hyperplane

The decision boundary learned by a Support Vector Machine, represented as a line in 2D space or a flat plane in 3D space.

Margin (SVM)

The space or distance surrounding the decision boundary line or hyperplane in a Support Vector Machine.

Support Vectors

The critical training data points closest to the boundary that determine the position and margin of the Support Vector Machine.

Feature Scaling

The data preparation process of transforming numeric attributes onto comparable numeric scales (e.g., adjusting word counts in thousands and link fractions between $0$ and $1$).

Kernel (SVM)

A function that provides boundary flexibility by projecting data into higher dimensions (e.g., an RBF kernel separating ring-shaped patterns).

SVM Parameter $C$

A regularization hyperparameter in Support Vector Machines that controls the penalty for margin violations; larger values penalize margin violations more heavily.

RBF Gamma

A hyperparameter for the Radial Basis Function kernel that determines the radius of influence of each training point, where larger values result in more localized influence.

K-Means Clustering

An unsupervised clustering algorithm where the user specifies $K$ centroids and points are grouped by proximity, best suited for compact, round clusters.

Hierarchical Clustering

An unsupervised clustering algorithm that builds nested groups within a tree structure, allowing inspection at both broad and fine granularities.

DBSCAN

A density-based clustering algorithm that connects dense neighborhoods of data points to find curved, irregular groups and separate noise points.

Principal Component Analysis (PCA)

A linear dimensionality reduction method that extracts linear components ranked by the amount of variation they capture, used for compression and noise reduction.

t-SNE

A nonlinear dimensionality reduction method that maps local similarities into 2D or 3D representations, primarily used for data exploration and visual inspection.

Forward Pass

The initial phase in neural network training where input features are passed through layers of neurons, weights, and biases to generate an output prediction.

Loss

A metric calculated during neural network training that quantifies the degree of disagreement or error between a model's prediction and the true target.

Backpropagation

The training step in neural networks that computes the mathematical gradients of the loss function with respect to each weight and bias.

Optimizer

The algorithm in neural network training that uses computed gradients to update and tune weights and biases to reduce loss.

Epoch

One complete pass of all training examples through the neural network during training.