CPSC 392 E1

Positive

Observation is positive

negative

observation is negative

True Postives

observation +, prediction +

False Postive

observation -, preditction +

True Negative

observation -, prediction -

false negative

observation +, predicition -

Accuracy

How close is predicitoin to the actual value

Accuracy Formula

(TN+TP)/N

Precision

How many selected values are relevant

Precision Formula

TP/(TP+FP)

Recall (Sensitivity)

How many relevant items are selected

Recall/Sensitivity/TPR Formula

TP/(TP+FN)

F1 Score Formula

2* ((recall*precision)/(recall+precision))

F1 Score

evaluates both precision and recall

Normalization

x-min(x)/max(x)-min(x)

Standardization

x-mean/Standard deviation

Overfit

When the model is too complex for the data given, resulting in reaching for correct values, think kNN but the dividing line is too specific

Underfit

when the model is too simple for the data given, resulting in undershooting the values. EX: using a linear model for non-linear data

Two ways of removing outliers

Removal, iterative Removal

Removal (Outliers)

Simply removing outliers

Iterative Removal (Outliers)

Removing and replacing outliers with values found from a model

2 ways of handling Missing Data

Deletion, imputation

deletion

Listwise, Pairwise, Variable Dropping

Imputation

LOCF, NOCB, interpolation, Extrapolation

Category Encoding

Giving categorical variables numeric values (Binary, Target-Based)

Binary Enconding

giving categorical variables a corresponding number (male 0, female 1)

Target Based Enconding

Binary encoding but finding the proportions

Supervised

Model is given a data set with labels

Unsupervised

Working with data without labels

Classification

Supervised Machine learning dealing with categorical value

Regression

Supervised Machine learning dealing with numerical value

algortithm

a methodical, logical rule or procedure that guarantees solving a particular problem. Contrasts with speedier heuristics

hyperparameter

the parameters that changes how the model deals with the normal parameters

FPR - False Positive Rate

FP/(FP+TN)

ROC

Graphs the different threshold points to see which threshold gives the most accurate data. y-axis: TPR, x-axis: FPR

AUC

area under the curve, used to determine which model's are more accurate. Whichever has the higher AUC value is more accurate

KNN

distance based classification to see which neighbors are closest, then out of the neighbors whichever is a majority one class is the class of the value

linear regression

a statistical method used to fit a linear model to a given data set