AI Fundamentals for Supply Chain & Machine Learning Flashcards

Flashcard practice set designed to review basic concepts of Artificial Intelligence, Machine Learning, algorithm types, evaluation metrics, and WEKA workflows for Supply Chain Management.

What is the relationship between Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL)?

AI is the broad field focused on making computers perform intelligent tasks. Machine Learning is a subset of AI where computers learn patterns from data. Deep Learning is a specialized type of Machine Learning that uses multi-layer neural networks.

How is Artificial Intelligence (AI) defined in the guide?

AI is the broad idea of building computer systems that can perform tasks associated with human intelligence, such as recognizing patterns, understanding language, making recommendations, solving problems, or making predictions.

How does Machine Learning (ML) differ from traditional rule-based programming?

Instead of a person writing every possible rule, a Machine Learning model learns relationships and patterns directly from examples and data.

What is Deep Learning, and what type of data is it especially suited for?

Deep Learning is a specialized branch of Machine Learning based on multi-layer neural networks. It is used to learn complex patterns in language, images, audio, and messy text documents.

What are the six beginner steps of how a Machine Learning model learns?

1. Give the computer examples. 2. Give it useful information. 3. Give it the known outcome during training. 4. The algorithm searches for patterns. 5. Test it on unseen data. 6. Evaluate the predictions using metrics like accuracy, precision, recall, F1, and a confusion matrix.

Why is it necessary to test a Machine Learning model on data it has not seen during training?

Testing on unseen data proves whether the model can generalize to new cases rather than just memorizing the training data.

In a spreadsheet-style dataset, what do rows and columns represent?

Rows represent individual cases (instances or observations), while columns represent information describing those cases (attributes or variables).

What is the difference between a feature and a target in a dataset?

A feature (input variable) is information given to the model to help it make predictions, whereas a target (output/class) is the specific outcome the model is tasked to predict.

How do numerical and categorical variables differ?

Numerical variables are represented by meaningful numbers that can be measured or counted (e.g., transit time = 45 days). Categorical variables represent discrete groups or labels (e.g., transport mode = Air / Sea / Road).

What is missing data, and why must it be addressed in Machine Learning?

A missing value is information that is absent for a particular observation. It must be noticed because missing data can negatively affect analysis and model training.

How do training data and testing data differ in function?

Training data is used by the algorithm to learn patterns (analogous to practice questions), whereas testing data is separate unseen data used to check model performance (analogous to an exam).

What is the primary difference between Supervised Learning and Unsupervised Learning?

Supervised Learning provides training examples with known correct target answers to learn input-output relationships. Unsupervised Learning has no target answer supplied and looks for inherent structures or groupings in the data.

What question does Classification answer, and what is a supply-chain example?

Classification asks 'Which category?'. A supply-chain example is predicting whether a shipment will be disrupted (Yes/No).

What question does Regression answer, and what is a supply-chain example?

Regression asks 'How much / how many?'. A supply-chain example is predicting how many days a shipment will be delayed (e.g., 7.4 days).

What question does Clustering answer, and what is a supply-chain example?

Clustering asks 'Which cases naturally group together?'. A supply-chain example is identifying which suppliers have similar risk profiles (Grouping into Cluster 1, Cluster 2, or Cluster 3).

What is the difference between an algorithm and a model?

An algorithm is the learning procedure or method (e.g., J48 Decision Tree, Logistic Regression), while a model is the specific pattern or set of rules generated after an algorithm has been trained on data.

How does the J48 Decision Tree algorithm work?

It learns a tree structure of decision rules, branching based on whether specific conditions are true or false to reach a prediction.

What task is Logistic Regression commonly used for despite having 'regression' in its name?

Despite its name, Logistic Regression is commonly used for classification tasks to estimate the probability of a categorical (often binary Yes/No) outcome.

How does Random Forest construct its predictions?

It combines multiple decision trees and aggregates their individual predictions to make a more robust overall prediction.

How does Naive Bayes estimate class probabilities?

It applies probability theory and Bayes' theorem under a simplifying assumption that input features are independent of one another.

How does 10-fold cross-validation evaluate a Machine Learning model?

The dataset is split into 10 equal parts. The model trains on 9 parts and tests on the remaining part. This process repeats 10 times so every part acts as a test set once, and the evaluation results are combined.

In a confusion matrix for shipment disruption, what are True Positives (TP) and False Positives (FP)?

True Positive (TP) occurs when a disruption is predicted and actually happens. False Positive (FP) occurs when a disruption is predicted but does not happen (a false alarm).

In a confusion matrix for shipment disruption, what are False Negatives (FN) and True Negatives (TN)?

False Negative (FN) occurs when no disruption is predicted but a disruption actually happens (a missed disruption). True Negative (TN) occurs when no disruption is predicted and no disruption happens.

What is the key difference between Precision and Recall in model evaluation?

Precision measures what proportion of positive warnings were actually correct (focusing on minimizing false alarms). Recall measures what proportion of all actual positive events were caught by the model (focusing on minimizing missed events).

What does the F1-score evaluate?

The F1-score provides a single numeric evaluation metric that balances both Precision and Recall.

What is Data Leakage, and what practical question helps detect it?

Data Leakage happens when information that should not be available at prediction time provides the model with unfair hints about the target. To detect it, ask: 'Would I genuinely know this information at the moment I want the prediction?'

What is the correct order of the mental workflow for evaluating a model in WEKA?

Dataset -> Preprocess -> Examine attributes -> Identify target/class -> Select classifier -> Choose evaluation method (e.g. 10-fold CV) -> Run -> Read confusion matrix/metrics -> Compare -> Interpret in Supply Chain terms.