Topic 3: Applications of Machine Learning in Marketing

Machine Learning (ML)

allows computers to learn patterns from data without being explicitly programmed.

Is ML a new concept?

No. It's older but has grown due to computing and data advancements.

ML vs. traditional programming

Traditional: rule-based; ML: learns rules from data.

Three main types of ML

Supervised, Unsupervised, Reinforcement learning.

Key difference between supervised and unsupervised learning

Supervised: labeled data; Unsupervised: unlabeled data.

Two types of supervised learning

Classification (categorical output), Regression (numeric output).

Reinforcement learning

Learning by interaction with feedback; long-term reward matters more.

Marketing examples for each ML type

Supervised: churn prediction; Unsupervised: customer segmentation; Reinforcement: ad bidding optimization.

Regression analysis vs. regression algorithm

Analysis: statistical method; Algorithm: ML-based, flexible.

Is linear regression both a regression analysis and algorithm?

Yes.

Is logistic regression both a regression analysis and algorithm?

Yes.

Artificial Intelligence (AI)

Broad field of machine-based intelligence, including ML.

Turing Test

Measures if a machine can mimic human behavior convincingly.

Deep Learning

Subset of ML using multilayered neural networks for complex tasks.

Layers in a neural network

Input, hidden (multiple), output layers. Deep = multiple hidden layers.

Data needs: traditional ML vs. deep learning

ML = small data; DL = large data needed for strong performance.

Relationship between AI, ML, DL

DL ⊂ ML ⊂ AI.

How can ML be used for customer segmentation?

Uses unsupervised learning (e.g., clustering) to group similar customers.

Key inputs of cluster analysis

Demographics, behavior, purchase history, engagement.

Key outputs of cluster analysis

Cluster groups, segment labels, centroids.

Steps in cluster analysis

Collect data, extract features, apply clustering, label segments, target segments.

Cluster centroid

The average profile of customers in a cluster; helps name the group.