MACHINE LEARNING

Question 1

Accuracy

Answer 1

How often model predicts correctly, e.g: 100 predictions, 80 is correct
80/100 = 0,8 0.8 or 80%

Question 2

Activation function

Answer 2

A function applied in neural netwroks to introduce non-linear transformations, enabling the network to model complex relationships. Exampes: Sigmoid, Relu

Question 3

Autoencoder

Answer 3

A type of neural network used for unsupervised learning, mainly for dimensionality reduction and feature learning by encoding inputs to a lower-dimensional space and reconstructing them.

Question 4

Back-propagation

Answer 4

An algorithm for traing neural networks, where errors are propagated backward through the network to adjust weights based on gradients.

Question 5

Bagging

Answer 5

“Bootstrap aggregating” - a technique to reduce variance by training multiple models on random subsets of the data and averaging predictions

Question 6

Basis function

Answer 6

Functions used in models to transform in put data to make it easier to fit linear models.

Question 7

Bias

Answer 7

The error due to overly simplistic models that do not capture patterns well.

Question 8

Bias-variance tradeoff

Answer 8

The balance between a models complexity(variance) and simplicity(bias) to achieve optimal performance

Question 9

boosting

Answer 9

An ensemble method: building models that correcting errors previos models made which leads to higher accuracy

Question 10

Bootstrap Sample

Answer 10

A sample obtained by randomly sampling with with replacement from dataset, often used in bagging and estimation
example:
- ball from a casket, and identify it with a color
- puts it back, and re-do the process
- point is you might pick same ball twice or more.

Question 11

Classification

Answer 11

A type of supervised learning where the goal is to predict discrete labels

Question 12

Clustering

Answer 12

An unsupervised learning technique for grouping similar data points toghether
- the machine finds patterns by itself
- netflix groups movies after genre

Question 13

Complete Linkage

Answer 13

A hierarchical clustering method where the distance between clusters is the maximum distance between any pair of points in the clusters
- biggest distance becomes “the distance between the clusters”

Question 14

Confusion Matrix

Answer 14

A table used to evaluate a classification model’s performance by comparing predicted and actual values.

Question 15

Cross-validation

Answer 15

A technique for assessing models:
- performance by dividing data into subsets,
- training on some
- testing on others to reduce overfitting

Question 16

Curse of dimensionality

Answer 16

• problem where data becomes sparse in high-dimensional spaces
• which makes it harder for models to find patterns

Question 17

Deep Learning

Answer 17

A subset of machine learning that uses neural networks with many layers to model complex data patterns
- uses many layers
- the more layers, the more complex things can the model learn

Question 18

Decision Boundary

Answer 18

The boundary that separates different classes in a classifier

Question 19

Classifier

Answer 19

Algorithm which is trained using labeled data
- employ math and statistics methods to generate predictions

Question 20

Decision Tree

Answer 20

A model that splits data based on feature values to predict the target label, forming a tree structure.

Question 21

Ensemble Methods

Answer 21

Techniques that combine multiple models to improve predictions, like bagging, boosting, and stacking

Question 22

Entropy

Answer 22

A measure of uncertainty or disorder in data, used in decision trees to determine the best splits

Question 23

Fairness

Answer 23

Ensuring that models do not discriminate or create biases against certain groups

Question 24

F1 Score

Answer 24

The harmonic mean of precision and recall, providing a balance between the two

Question 25

Feature Extraction

Answer 25

Deriving new features from raw data -often for dimensionality reductin - or improved model performance

Question 26

Feature Selection

Answer 26

choosing the most relevant features for a model to improve accuracy and reduce complexity

Question 27

Generalization

Answer 27

Ability for a model perform well on unseen data - bad generalization: Can 2+2=4, but cant do 3+3=6 - or, only bicycle on one spesific cycle

Question 28

Gradient Descent

Answer 28

An optimization algorithm to minimize a model's loss function by iteratively adjusting parameters - Adjusting a model until it gets perfect

Question 29

Hierarchical Clustering

Answer 29

A clustering method hat creates a tree of clusters by iteratively merging or splitting clusters

Question 30

Imbalanced data

Answer 30

A dataset where some classes are more frequent than others, often requiring techniques like resampling

Question 31

Imputation

Answer 31

Filling in missing data - commonly done using statistical - or model-based techniques

Question 32

Inductive basis

Answer 32

Assumptions to generalize beyond training data

Question 33

Information gain

Answer 33

metric used in decision trees to measure the reduction in entropy after a split

Question 34

Kernel

Answer 34

A function that transforms data into a higher-dimensional space pros: - used in SVM to handle non-linearly separable data

Question 35

K-means

Answer 35

A clustering algorithm that partitions data into k clusters by minimizing the distance of points from the cluster centroid - You want to centralize the clustuers, by doing the same thing multiple times -sensitive for: startposition, outliers, scale on features

Question 36

k-NN

Answer 36

A classification algorithm that assigns labels based on the majority label among the k closest points -

Question 37

Learning rate

Answer 37

Steps a model takes during gradient descent

Question 38

linear regression

Answer 38

A model that finds a linear relationship between a dependent variable and one or more independent variables

Question 39

Lloyds algorithm

Answer 39

A common method for k-means clustering, iterating between assigning points to clusters and updating centroids

Question 40

Logistic regression

Answer 40

A model used for binary classification by modeling the probability of an outcome with a sigmoid function

Question 41

loss function

Answer 41

function that finds error from a models predicts - Used to guide training

Question 42

MAE(mean absolute error)

Answer 42

average of absolute difference between predicted and actual values

Question 43

Missing data(MCAR, MAR, MNAR)

Answer 43

Types of missing data MCAR- completely random MAR- related to observed data MNAR- related to missing data itself

Question 44

Model Evaluation

Answer 44

Process of assessing a model's performance using metrics like accuracy, precision, recall and RMSE

Question 45

Model Selection

Answer 45

The process of choosing the best model among various options based on performance

Question 46

Naive bayes

Answer 46

Probabilistic classifier based on Bayes theorem, assuming features are independent

Question 47

Neural Network

Answer 47

Model inspired by the human brain, consisting of layers of neurons that process data for tasks like classification and regression

Question 48

No free lunch theorem

Answer 48

No model works best for all problems

Question 49

overfitting

Answer 49

When a model learns noise in the training data, performing poorly on new data

Question 50

Oversampling

Answer 50

A technique to balance classes by replicating or generating new samples of the minority class

Question 51

Principal Component Analysis (PCA)

Answer 51

A dimensionality reduction technique that projects data into a lower-dimensional space while retaining variance

Question 52

Pruning

Answer 52

Process of removing parts from a decision tree to prevent overfitting

Question 53

Radial Basis function

Answer 53

A function used in machine learning, often in neural networks and SVMs, to handle non-linear patterns

Question 54

Random Forest

Answer 54

An ensemble of decision trees that improves accuracy and reduces overfitting by aggregating predictions

Question 55

Recall

Answer 55

Ratio true positive predictions to the total actual positives

Question 56

Rectified linear Unit (ReLU)

Answer 56

An activation function that outputs zero for negative inputs and the input itself for positive values