classification Flashcards
features
measurable quantities
discriminatory power
quality of a feature that can help us classify
feature space
the set of all features from the feature vector
classification
process of organising things into groups or classes
machine learning
the scienec of making
features for image classification
form factor, Euler number
What does a negative Euler number indicate?
number of holes is greater than the number of objects
typical features of sound classification
waveform, frequency spectrum, formants(peaks in the spectrum), spectrogram, MFCC
binarisation
converting an image or a signal into a binary format, typically consisting of only two intensity values, often black and white or 0 and 1
rule based classifier should have…
mutually exclusive rules and those ruler should be exhaustive
supervised classification
+
examples
relies on having a set of examples
template matching, decision tree, neural networks, naive Bayes
unsupervised classification
doesn’t have a training set
clustering
fingerprint
a feature that can be compared
perceptron
a structure that contains a neuron with adjustable weights
neural network
a multi-layer perceptron
Hidden Markov Models
probabilistic reasoning algorithm that works on a set of temporal data
reinforcement learning
between supervised and unsupervised
classification error
error rate
num of errors / num of attempts
test set
has positive and negative instances
true positive
an instance of the positive class that is correctly classifies
true negative
an instance of the negative class that is correctly rejected
false positive
an instance of the negative class that is wrongly classified
false negative
an instance of the positive class that is wrongly rejected
dangerous!!
accuracy
a=(tp+tn)/n
precision
p=tp/(tp+fp)
recall
r=tp/(tp+fn)
confusion matrix
gives info on how frequently instances of class X are correctly classified as X or misclassified
components of a classification system
sensing module, preprocessing mechanism, feature extraction mechanism, classifier, training set if needed
