Big Data and Deep Learning for Cybersecurity - Introduction to Machine Learning & Data Mining

Vocabulary flashcards covering core definitions and concepts from the Big Data and Deep Learning for Cybersecurity lecture, including Big Data characteristics, supervised and unsupervised mining tasks, generative versus discriminative modeling, the KDD process, and the CRISP-DM lifecycle.

Data Mining

The use of efficient techniques for the analysis of very large collections of data and the extraction of useful and possibly unexpected patterns in data (knowledge discovery from data).

Network Intrusion Detection

The identification of suspicious activity in a computer network, such as scanning, attempted unauthorised access, or unusual data transfers.

Phishing Email Detection

The identification of deceptive emails that attempt to steal credentials, obtain sensitive information, or persuade users to perform harmful actions.

Volume (Big Data)

A Big Data characteristic referring to the total scale or amount of data, such as millions of medical images or billions of network events.

Velocity (Big Data)

A Big Data characteristic describing how quickly data arrives and must be processed, such as login events continuously generated across an organisation.

Variety (Big Data)

A Big Data characteristic describing the different formats and sources of data, including structured tables, CT images, clinical text, device logs, and videos.

Veracity (Big Data)

A Big Data characteristic defining data quality and trustworthiness, addressing problems like missing values, incorrect labels, or unreliable measurements.

Value (Big Data)

A Big Data characteristic denoting the useful outcome obtained from data, such as better segmentation, earlier threat detection, or reduced costs.

Supervised Learning (Data Mining)

A goal-oriented data mining method whose purpose is to explain or predict the value of a selected target field in terms of all other fields based on a hypothesis, executing tasks such as classification and prediction.

Unsupervised Learning (Explorative Methods)

Data mining methods where no target field is specified, aimed at discovering interesting patterns or structures in data through tasks like clustering and association rules (affinity grouping).

Prediction Methods

Data mining tasks that use some variables to predict unknown or future values of other variables.

Description Methods

Data mining tasks that find human-interpretable patterns that describe the underlying data.

Inductive Modeling

The learning objective of developing a general model or hypothesis from specific examples, encompassing function approximation (curve fitting) and classification (concept learning, pattern recognition).

Classification

A predictive modeling task that finds a model for a class attribute as a function of the values of other attributes to predict discrete categories for data instances.

Discriminative Model

A machine learning model that seeks to draw classification boundaries between different categories in the feature space.

Generative Model

A machine learning model trained to model the spatial distribution of data and capture structural correlations (e.g., that objects resembling boats are likely to appear close to water) in order to generate new data instances.

Clustering

An unsupervised technique of finding groups of objects such that intra-cluster distances are minimized (objects within a group are similar) and inter-cluster distances are maximized (objects across different groups are dissimilar).

Market Segmentation

A clustering application aimed at subdividing a market into distinct subsets of customers based on geographic and lifestyle information so that any subset can be targeted with a distinct marketing mix.

Document Clustering

A clustering application aimed at finding groups of similar documents based on the frequencies of frequently occurring terms within them.

Objective Interestingness Measures

Pattern evaluation metrics based strictly on statistics and the structural properties of patterns, such as support and confidence.

Subjective Interestingness Measures

Pattern evaluation metrics based on a user's beliefs regarding the data, assessing characteristics such as unexpectedness and novelty.

KDD Process

Knowledge Discovery in Databases; an iterative process consisting of data consolidation and warehousing, selection and preprocessing, data mining, and interpretation and evaluation.

CRISP-DM

CRoss Industry Standard Process for Data Mining; a standardized, repeatable six-phase process framework for planning, executing, and recording data mining projects.

Business Understanding (CRISP-DM)

The initial CRISP-DM phase focused on determining business objectives, assessing the situation, translating business goals into technical data mining goals, and producing a project plan.

Data Understanding (CRISP-DM)

The CRISP-DM phase that starts with initial data collection and proceeds to describing, exploring attribute distributions, and verifying data quality to detect anomalies and form hypotheses.

Data Preparation (CRISP-DM)

The CRISP-DM phase covering table, record, and attribute selection, data cleaning, data construction, data integration, and data formatting, typically consuming over $90\%$ of the time.

Modeling (CRISP-DM)

The CRISP-DM phase where specific modeling techniques are selected, test designs are generated (e.g., train-test splits), models are built, and their parameters are calibrated.

Evaluation (CRISP-DM)

The CRISP-DM phase that assesses the degree to which built models satisfy original business objectives, reviews the process to detect overlooked issues, and decides on next steps.

Deployment (CRISP-DM)

The final CRISP-DM phase where the discovered knowledge and models are organized into a plan, monitored, maintained, reported, and integrated into repeatable operational use.