Predictive Analytics: Goals and Variables

Learning Goals

• Understand what prediction is
• Identify predictor and response variables
• Discuss details about data used for advanced analytics
• Discern types of predictive models
• Identify differences between ML and AI
• Explain the two types of AI
• Ability to come up with advanced analytics questions

Topics Covered

• Prediction as a concept
• Transitioning predictions into decision-making
• Predictor and response variables
• Data used in advanced analytics
• Types of predictive models
• Crafting advanced analytics questions
• Machine Learning (ML) and Artificial Intelligence (AI), and the two types of AI

Types of Analytics

• Descriptive Analytics: Examines data to answer "What happened?" or "What is happening?"
• Diagnostic Analytics: Examines data/content to answer "Why did it happen?"
• Predictive Analytics: Examines data/content to answer "What is going to happen?" or "What is likely to happen?"
• Prescriptive Analytics: Examines data/content to answer "What should be done?" or "What can we do to make X happen?"

Prediction

• Prediction uses existing information to generate new information.
• Prediction: Foretelling based on observation, experience, or scientific reason.
• In science, a prediction is a rigorous, often quantitative statement forecasting events under specific conditions.

Predictive Modelling

• Aids in solving business problems by improving efficiency, making better decisions, and enabling innovation.
• Business problems are reframed as prediction problems.
• Prediction questions include:
  • "What is someone likely to do in a given scenario?"
  • "What is likely to happen if...?"
  • "How much is expected when...?"
• Analytics aims uncover insights that give the businessperson knowledge in order to make informed decisions.
• Predictive modeling examples include:
  • Pipe breakdown prediction
  • Survival analysis: Predicts the occurrence and timing of events like customer churn, patient survival, and loan defaults.
  • Network analysis: Targets specific individuals within a network for advertising.
  • Net lift response modelling: Predicts the likelihood of customer response to a promotion/campaign.

Prediction vs. Decision

• Decisions can be:
  • Automated with scores derived from models
  • Partially automated for standard behaviors with override rules for exceptions, using the model for predictions in most cases
  • Not automated: Requires human judgment using model scores and predictions.
• The first step is to make a prediction, followed by a decision based on that prediction.
• The maturity of analytics depends on the automation level of the decision process and the amount of human intervention needed.

Predictor and Response Variables

• Model: A mathematical representation of an object or process.
• Predictive Modeling: Employs methods to develop models examining datasets for patterns, calculating outcome probabilities.
• The relationship between predictor and behavioral (response) data is key to predictive modelling.
• Predictive analytics: Analyzes data to see how predictor data differentiates behaviors, so it that they can predict outcomes.

Variables Defined

• Predictor Variable: Variable to predict the response, also known as the independent variable, feature, attribute, or X-variable.
• Response Variable: Variable the mode aims to predict, also known as the dependent variable outcome, target, or Y-variable.

Data For Predictve Analtyics

• Sources of predictor variables include:
  • Existing metadata
  • Current decision-making process
  • Input from subject-matter experts in operational areas
  • External data sources
  • Research and peer experience
• Finlay (2014) wrote “Predictive Analytics, Data Mining and Big Data (Business in the Digital Economy)"
• Data types with corresponding data descriptions:
  • Primary behaviors: Past behaviors of the same type as the behavior being predicted (e.g., past burglaries).
  • Secondary behaviors: Past behaviors similar to the one being predicted (e.g., minor crimes).
  • Tertiary behaviors: Past behaviors with no obvious link to what's being predicted (e.g., selling items on eBay).
    • Geo-demographic: Information such as age, income, occupation, and education level.
    • Associate data: Information about a person's connections and their behaviors (e.g., friends who committed a burglary).
    • Sentiments: Attitudes, feelings, and opinions (e.g., social media likes/dislikes, survey answers).
    • Network: The nature of a person's connections (e.g., friends with criminal offenses).

Machine Learning

• Supervised learning:
  • It predicts outcomes using labeled data.
  • Learns patterns from labeled data in a "training set" where correct answers or "labels" are known.
  • Uses classification models to predict event probabilities.
  • Regression models predict a quantity or estimate value.
• Unsupervised learning:
  • Predicts outcomes using unlabeled data.
  • Models underlying data structure/distribution, to learn more about it.
  • Data is unlabeled (i.e., correct answer is unknown).
  • Clustering finds a structure or pattern in a collection of uncategorized data.
  • Association discovers relationships between variables.

Comparing Supervised and Unsupervised Machine Learning

• Supervised:
  • Involves algorithms trained using labeled data.
  • Is computationally simple.
  • Is more accurate and higher trustworthiness overall.
• Unsupervised:
• Involves algorithms that make assumptions based on unlabeled data.
  • Is computationally complex.
  • It can be less accurate and lower trustworthiness overall.

Predictive Analytics Questions

• Questions:
  • "What is someone likely to do in a given scenario?"
  • "What is likely to happen if...?"
  • "How much is expected when...?"
  • "What is the expected for...?"
  • "What is the likelihood that someone will choose...if...?"
  • "What is the probability that...belongs to...category?"

Machine Learning vs Artifical Intelligence

• AI: Computer systems can perform tasks normally requiring human intelligence (e.g., visual perception, speech recognition).
• ML: An AI discipline this allows computers to handle new situations via analysis, self-training, and observation.
• Deep Learning (DL): Is an ML subfield using artificial neural networks.

Types of AI

• Narrow AI: For a narrowly defined or single task.
• General AI: It can handle a broad set of tasks comparable to human capabilities.

Narrow vs General AI

• Narrow AI:
  • Goal: It is intended to do a single task.
  • Principle: Process and analyze data, and build prediction models to make quantitative predictions.
  • Solves problems that are framed as prediction problems.
  • Aka Weak AI
  • Includes all AI that exists today
• General AI:
  • Goal: It is intended to do a broad set of tasks that a human is capable of.
  • Principle: Replicate/imitate human cognitive processes.
  • Intended to solve any problem that a human can solve
  • Aka Strong / Full AI
  • Found in fiction books and research phases.

Analytics Applications Examples

• Refer to narrowly defined AI:
  • Recommender systems
  • Targeted advertising
  • Speech recognition
  • Typing autocomplete
  • Gaming
  • Fraud and risk detection
  • Logistics
  • Sports cameras

Explore predictive analytics, its main goals concerning future events, and the types of questions it addresses. Understand the roles of predictor and response variables in predictive modeling, and how existing metadata can be a valuable source for predictor variables.