UNIVERSITY OF OXFORD | MSC AI
FOR BUSINESS
MODULE: CLASSICAL MACHINE LEARNING (CML)
TERM 1 LECTURE NOTES: FOUNDATIONAL ALGORITHMS & PREDICTIVE
MODELING
1. THE "CLASSICAL" VS. "DEEP" DISTINCTION
In the Oxford curriculum, Classical ML refers to algorithms that learn from
structured data (rows and columns) rather than unstructured data (pixels/audio).
• Key Advantage: Interpretability. In business, you often need to explain why
a customer was denied a loan. Classical models allow for this; Deep
Learning often does not.
• The "No Free Lunch" Theorem: No single algorithm works best for every
problem. A leader's job is to match the algorithm to the data.
2. REGRESSION: PREDICTING CONTINUOUS VALUES
Regression is the most common tool for Forecasting (Sales, Prices, Demand).
2.1 Linear Regression
The simplest model that establishes a relationship between a dependent variable
(Y) and one or more independent variables (X).
• Equation:
• Business Use: Predicting next quarter's revenue based on current
marketing spend.
• Assumption: It assumes a straight-line relationship. If the data is curved,
the model fails.
FOR BUSINESS
MODULE: CLASSICAL MACHINE LEARNING (CML)
TERM 1 LECTURE NOTES: FOUNDATIONAL ALGORITHMS & PREDICTIVE
MODELING
1. THE "CLASSICAL" VS. "DEEP" DISTINCTION
In the Oxford curriculum, Classical ML refers to algorithms that learn from
structured data (rows and columns) rather than unstructured data (pixels/audio).
• Key Advantage: Interpretability. In business, you often need to explain why
a customer was denied a loan. Classical models allow for this; Deep
Learning often does not.
• The "No Free Lunch" Theorem: No single algorithm works best for every
problem. A leader's job is to match the algorithm to the data.
2. REGRESSION: PREDICTING CONTINUOUS VALUES
Regression is the most common tool for Forecasting (Sales, Prices, Demand).
2.1 Linear Regression
The simplest model that establishes a relationship between a dependent variable
(Y) and one or more independent variables (X).
• Equation:
• Business Use: Predicting next quarter's revenue based on current
marketing spend.
• Assumption: It assumes a straight-line relationship. If the data is curved,
the model fails.
