Page 343 - AI Ver 3.0 Class 11
P. 343

Brainy Fact

                      One  of  the  first  algorithms  used  for  machine  learning  was  the  Naive  Bayes  classifier.  Spam
                      filtering systems used Naive Bayes till 2010.





                 Types of Classification

                 Classification is a supervised learning concept which groups a set of data into classes. It is mainly of four types, which
                 are as follows:

                 •  Binary classification: It refers to classification problems/tasks that have
                    only two class labels. For example, email spam detection (spam or not),
                    churn  prediction  (whether  customers  will  stop  doing  business  on  a
                    particular website or he will continue). In general, binary classification   x 2
                    tasks involve 2 labels—normal and abnormal. For example, “no-spam”
                    is  a  normal  condition  and  “spam”  is  an  abnormal  condition.  Another
                    example is that “tumour not detected” is a normal state and “tumour
                    detected” is an abnormal condition.

                                                                                                          x 1
                                                       •    Multiclass  classification:  It  implies  those  classification  tasks  that
                                                            have more than two class labels. Each entity is assigned to one class
                                                            without any overlap. For example, face classification, animal species
                                                            classification,  optical  character  recognition.  In  contrast  to  binary
                 x 2                                        classification, multiclass classification does not have the concept of
                                                            normal and abnormal classes. Instead, the examples are classified as
                                                            belonging to one of the several known classes.

                                                              The  number  of  classes  may  be  very  large  in  some  problems.  For
                                                            example,  a  model  may  tag  a  photo  as  belonging  to  one  among
                                                            thousands  of  faces  in  a  face  recognition  system.  Text  translation
                                    x 1                     models are also a special type of multiclass classification.




                 •  Multi-label  classification:  Multi-label  classification  is  used  when  a
                    situation might belong to more than one class at the same time. This
                    implies that for a given input, the output may contain a collection of
                    labels  instead  of  a  single  one.  For  example,  document  classification   x 2
                    (where a document may be classified into various groups at the same
                    time,  such  as  “science”  and  “technology”),  and  object  detection  in
                    images etc.



                                                                                                         x 1




                                                                                  Machine Learning Algorithms   341
   338   339   340   341   342   343   344   345   346   347   348