BSNS112 - Non Linear revision

Why is data quality important when interpreting business data?

it is a major issue in interpreting business data.

What are the characteristics of good quality data?

It has structure and order, variables will be labelled, clearly typed, and (ideally) have no missing values

What is metadata, and why is it important?

It describes who created the data, why they created it, what it can be used for, the structure of the data, and so on.

What does Principal Components Analysis (PCA) do with the variation in data?

It re-orients variation along more meaningful axes.

What does each principal component represent in PCA?

Each describes a major source of variation; the largest source of variation in the data lies within the first principal component.

What is the relationship between principal components?

They are orthogonal and uncorrelated.

Why do we transform variables?

To make them more meaningful

Why do we standardise variables?

To remove the effect of scale differences.

What is the purpose of cross-validation?

It estimates a model's performance on unseen data.

How can we encode categorical data numerically?

Using binary or dummy encoding.

What is logistic regression used for?

Models binary outcomes as the log odds of the event against features.

How do decision trees make predictions?

Predict unknown responses using yes/no tests on predictors.

What happens when a decision tree splits data?

Splitting data with yes/no tests reduces variance in subsets.

How can we prevent decision trees from overfitting?

Cross-validation helps control the size of decision trees.

What is a benefit of visualising decision trees?

Provides insights.

What are neural networks made of?

Consist of layers of perceptrons (computational neurons).

How do perceptrons work?

They compute weighted sums of inputs and apply nonlinear activation functions.

How are weights determined in neural networks?

They are randomly initialised and refined through supervised learning and backpropagation.

How can cross-validation be used with neural networks?

To select the number of nodes in a neural network.

Why are neural networks important?

They are flexible and foundational for modern AI methods.

What is the goal of unsupervised learning?

Identifies relationships between instances to form clusters.

What do clustering methods use to determine similarity?

It relies on similarity measures like Euclidean distance.

Why is standardising important in clustering?

It ensures variables contribute equally.

What is k-Means clustering?

Used to determine a specified number of clusters using an iterative algorithm.

What is a dendrogram?

Builds clusters hierarchically based on similarity.

How are clusters formed from a dendrogram?

By cutting the dendrogram at specific levels.