Predictive Analytics: Techniques and Applications

Introduction to Analytics (2022-2023) Lesson 4: Advanced Analytics, Machine Learning & AI

• Objectives: Understand the meaning of prediction, recognize predictor and response variables, discuss data used in advanced analytics, distinguish types of predictive models, differentiate between ML and AI, explain the two types of AI, and formulate advanced analytics questions.
• Agenda: Define prediction, transition from prediction to decision, identify predictor and response variables, discuss data used in advanced analytics, classify types of predictive models, ask advanced analytics questions, differentiate ML and AI including the two types of AI, and introduce the group assignment.

Types of Advanced Analytics

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

Prediction

• Prediction: Using existing information to generate new, unknown information.
• Predict: Foretelling based on observation, experience, or scientific reasoning.
• Scientific Prediction: A rigorous, quantitative statement that forecasts outcomes under specific conditions.

Predictive Modeling

• Predictive modeling helps improve efficiency, enable new organizational capabilities and make better decisions.
• Predictive modeling involves reframing a business problem as a prediction problem.
• Typical Prediction Problems: "What is someone likely to do in a given scenario?", "What is likely to happen if...?", and "How much is expected when?".
• Analytics identifies insights to provide businesspeople with the knowledge to make informed decisions.

Predictive Modeling Examples

• Survival Analysis: A model that predict timing and the occurrence of events – customer churn, patient survival.
• Network Analysis: Targets individuals within a network for advertising to optimize budget.
• Net Lift Response Modeling: Predicts the likelihood of customers responding to a promotion or campaign.
• Decision-making can be automated, partially automated, or not automated, based on model scores and human judgment.

Prediction vs Decision

• Automated decisions are based on model scores, partially automated decisions allow overrides for exceptional cases, but not fully automated decisions require human judgment.
• Analytics progression: data -> analytics -> decision support -> action.
• The extent of automation and human intervention dictate analytics maturity.

Predictor and Response Variables

• Model: a mathematical representation of an object or process.
• Predictive Modeling: Use of mathematical and computational methods to create predictive models to analyze datasets, patterns, and probabilities.
• Predictive Modeling Focus: aims to understand relationships between predictor and behavioral data.
• Predictive analysis identifies patterns in data allowing for differentiation between behaviors and predicting outcomes.

Definitions of Predictor and Response Variables

• Predictor Variable: Used to predict the outcome of the response - independent, feature, attribute and X-variable.
• Response Variable: The variable that the model aims to predict - dependent, outcome, target and Y-variable.
• Practice: Streaming platform data used to improve movie recommendations.
• The Goal: Identify the platform movie recommendations algorithm.

Sources of Predictor Variables

• Existing Metadata: Existing documentation of business data and constraints.
• Current Decision-Making Process: What data is used to make current business decisions.
• Subject Matter Experts: People involved in day-to-day business processes know important and relevant information.
• External Data Sources: Outside entities with relevant knowledge, like credit rating agencies.
• Research/Peer Experience: Data used by your industry, competitors, and peers.

Data Used for Advanced Analytics

• Primary Behaviors: Past behaviors of the same type you want to predict (e.g. previous burglaries).
• Secondary Behaviors: Past behaviors that are similar but different to the one you want to predict (e.g. minor crimes).
• Tertiary Behaviors: Past behaviors without an obvious connection to the behavior you want to predict (e.g. selling items on eBay).
• Geo-Demographic: Information about a person's background (age, income, occupation).
• Associate Data: Information about a person's connections and their behaviors (e.g. friends who committed a crime).
• Sentiments: a Person's attitudes, feelings, and opinions.
• Network: Information about the nature of connections between a person and their associates.

Machine Learning

• Supervised Learning: Predicts outcomes using labeled data to detect patterns, and response variables.
• Unsupervised Learning: Models the underlying structure or distribution in unlabeled data to discover patterns. Lacking in the correct answers.

Supervised vs. Unsupervised Learning

• Supervised machine learning trains algorithms using labeled data.
• Unsupervised learning applies algorithms to unlabeled data which makes it all the more complex and untrustworthy, but simpler.

Supervised Learning: Predictive Models

• Classification Problem: Predicting the probability of an event from a list of possibilities and is categorical.
• Regression problem: Find the function that fits the data to estimate a numerical value.

Asking Advanced Analytics 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 metric for...?
• What is the likelihood that someone will choose if...?
• What is the probability that belongs to category?

Machine Learning vs AI (Artificial Intelligence)

• AI: Computer systems performing tasks requiring human intelligence (visual perception, speech recognition, decision-making, language translation).
• ML Discipline: Artificial intelligence that enables computers to adapt via self-training and analysis.
• Deep Learning: A subset of machine learning using artificial neural networks inspired by the structure of the brain.
• Scope: Computer Science > AI > Machine Learning > Deep Learning.

AI Types

• Narrow AI: Designed for single tasks.
• General AI: Capable of broad tasks comparable to human capabilities.
• Narrow Al Goal: Designed for a single task and build quantified prediction models.
• General AI: Replicate or imitate human cognitive process.
• Narrow AI's problems are based on prediction problems only, whereas General AI can solve any problems that a human can solve.

Analytics Applications Examples – Narrow AI

• Recommender Systems: Based on user's previous search results.
• Targeted Advertising: Based on user behavior and social media analytics.
• Speech Recognition:: Use sound patterns to identify the best match to predict intention.
• Typing Autocomplete: Predicts words based on the context and the first letters.
• Gaming: Algorithms improves and upgrade themselves as the Player goes through levels.
• Fraud and Risk Detection: Predicts likelihood to default on loan by using customer profiles.
• Logistics: Aims to find best routes to ship the best time to deliver, the best mode of transport.
• Sports Cameras: Use of real time predictions in-game directions.

Explore predictive analytics through questions covering its goals, techniques, and applications. Understand different types of analytics and their role in decision-making. Learn how reframing business problems as prediction problems can improve outcomes.