Practical Machine Learning

is a single statistic that captures the performance of a dichotomous diagnostic test. Informedness is its generalization to the multiclass case and estimates the probability of an informed decision.

A measure of how closely two judges agree on classification, taking into account the probability of judges agreeing via random chance.

Ratio of True Positives divided by (True Positives + False Negatives) TP/(TP + FN)

Ratio of True Negatives divided by (False Positives over True Negatives) TN/(TN + FP))

Ratio of True Positives divided by (True Positives plus False Positives) TP/(TP+FP)

Ratio of True Negatives divided by (True Negatives plus False Negatives) TN/(TN+FN)

Ratio of True Positives plus True Negatives divided by all positives and negatives (TP+TN)/(TP+TN+FP+FN)

Area under the Curve (For example see Receiver Operating Characteristics in Wikipedia.)

Receiver Operating Characteristics curve (see Wikipedia)

sometimes called rotation estimation, is a model validation technique for assessing how the results of a statistical analysis will generalize to an independent data set.

Successive steps in assessing how well a machine learning algorithm performs on specific cases, improving the algorithm and then finally testing it against cases that WERE NOT USED TO TRAIN IT.

A method for selecting test data versus training data

Break data into K equal size subsets.

Use 1 for testing and K-1 for training. Test each individual subset by training with it's K-1 counterparts.

Literally leave one sample out of the training set and then test the prediction of that sample. Iterate by training and then testing on each sample, leaving out a different sample each iteration.

Random sampling with replacement. Underestimates error.

".632 bootstrap" is designed to try to compensate for that underestimation. See Wikipedia article on Cross Validation.

Generalized Linear Model

A statistical test used on paired nominal data. It is applied to 2 × 2 contingency tables with a dichotomous trait, with matched pairs of subjects, to determine whether the row and column marginal frequencies are equal (that is, whether there is "marginal homogeneity").

Linear discriminant analysis, Regression, Naive Bayes, Support Vector Machines, Classification and regression trees, Random Forests, Boosting, etc.

The number of cases of a type that are present in a particular population at a given time

The number of new cases that develop in a given period of time (disease, for example).

is the generalization of Youden's J statistic to the multiclass case and estimates the probability of an informed decision.

?

?

Leave one out cross validation

?

repeated cross validation

?

improving a model to better predict

is a prototype-based supervised classification algorithm. LVQ is the supervised counterpart of vector quantization systems. It is a special case of an artificial neural network, more precisely, it applies a winner-take-all Hebbian learning-based approach.

It is a precursor to self-organizing maps (SOM) and related to neural gas, and to the k-Nearest Neighbor algorithm (k-NN). LVQ was invented by Teuvo Kohonen

bias of a learning algorithm is the set of assumptions that the learner uses to predict outputs given inputs that it has not encountered

inferring a function from labeled training data. The training data consist of a set of training examples. Each example is a pair consisting of an input object (typically a vector) and a desired output value (also called the supervisory signal).