Class 12: Perform & Communicate Predictive Analysis (Machine Learning)

What is Generative AI?

Can create new content or data, such as text, images, or music, by learning from existing datasets

What does AI encompass?

• Machine Learning, Computer vision, etc

• Generative AI

• Large Language Model

What is Machine Learning?

• Type of artificial intelligence

• The ability of a computer (or machine) to automatically learn on its own without being explicitly programmed to do so

What are some uses of Machine Learning?

• Internet search engines

• Email spam filters

• Personalized recommendations

• Identify unsuual banking transactions

• Voice recognition

How does the artificial intelligence algorithm learn?

By accessing data (the more data, the more the machine can learn)

What do Supervised Algorithms do?

Analyze both independent variables (inputs) and dependent variables (outputs)

What do Supervised Algorithms use the output measure for?

To train or “supervise” algorithms to classify/predict accurately

What is an example of supervised machine learning?

Decision Tree

What will most accounting analytics involve?

Supervised ML

What does Unsupervised Machine Learning do?

Analyzes only independent variables (inputs)

What do algorithms of unsupervised machine learning do?

Uncover any hidden patterns without any “supervision” to classify/predict accurately

What is Unsupervised Machine Learning used in?

Music prediction algorithms

What is an example of Unsupervised Machine Learning?

Principal Components Analysis (PCA)

What is Principal Components Analysis?

• Identifies common components of underlying data

• Dimension reduction technique (1 or 2 dimensions)

What can we apply machine learning models to?

To making out of sample predictions

How can you apply the ML model to out of sample data to generate predictions?

Can do this in a single dataset by splitting the dataset (one for testing and one for training)

What can be a huge problem for machine learning?

Overfitting

What happens in Under-Fitting?

Too simple to explain the variance

What happens in Over-Fitting?

It models the noise in the training data instead of the actual pattern, leading to poor generalization on unseen data

What are Decision Tree Algorithms?

• Allows for nonlinearity, but prone to overfitting

• Base of other ML models

• Can be used for both classification and regression

What are the decision tree types?

• Outcome is a classification = classification tree

• Outcome is a number = regression tree

What does each node contain in a decision tree?

A decision that can be answered using an observation’s data

What is Gini Impurity?

• Measures how pure a leaf is

• “Pure Leaf” = All observations on leaf have the same outcome value

• Use to identify which variable and decision to place in each node

What do you do in a decision tree if the condition is true?

Move to the left (if false, then move right)

What does a Linear Regression minimize?

The sum of squared errors

Do machine learning trees have the constraint of minimizing the sum of squared errors?

No, which means it can fit data better (but it could also overfit)

What is a Random Forest?

A machine learning model that uses multiple decision trees to improve accuracy and control overfitting by averaging their predictions

What is the difficult part of supervised machine learning?

Understanding how to use and interpret the model:

• When to use each model to answer your accounting analytics question

• What the model actually does

• How to interpret the model’s output

What is Robotic Process Automation (RPA)?

Another form of machine learning in accounting that is a software program that automates certain repeatable tasks

What are the pros of RPA?

Eliminates human error/bias, can do repetitive tasks very quickly

What are the cons of RPA?

Only does exactly what it is programmed to do