Page 318 - AI Ver 1.0 Class 10
P. 318

In the above Scenario of Board Exams prediction:
              If the model always predicts All Positives that there will ALWAYS BE A BOARD EXAM irrespective of the reality.
              It  would  take  into  consideration  all  the  Positive  conditions,  which  are  True  Positive  (Prediction  =  Yes  and
              Reality = Yes) and False Positive (Prediction = Yes and Reality = No). Here students would always be on their
              toes and vigilant to find out if the Board Exams will happen or not, and keep verifying if the prediction is TRUE
              or FALSE.
              Importantly, if the Precision is less, that is, if there are more False Predictions, the students might become laid back,
              and might not check it more often, considering that the board exams will not happen. That is why Precision of the
              model is an important aspect for evaluation. So, if the Precision is more, that would mean that False Positive cases
              are less than the True Positive cases.
              So if the model is 100% precise, that would mean that whenever the model says there are exams happening (True
              Positive), the exams would definitely happen. There can be rare exceptional situations where the model would not
              be able to predict the exams, but the exams are there (False Negative). In this case the Precision value does not
              get affected, as the False Negative is not considered by the model for the evaluation. Which raises a question: Is
              Precision a good parameter for performance of the model?


                            Task



                 Now, find out is good Precision equivalent to a good model performance?
                 _______________________________________________________________________________________________________________________________

                 _______________________________________________________________________________________________________________________________

                 _______________________________________________________________________________________________________________________________

                 _______________________________________________________________________________________________________________________________

                 _______________________________________________________________________________________________________________________________





              Recall/Sensitivity

              Recall is defined as the fraction of positive cases that are correctly identified. It majorly takes into account the true
              reality cases i.e.; it is a measure of our model correctly identifying True Positives. Its formula is:


                                                                 True Positive
                                                 Recall =
                                                         True Positive + False Negative
                                                            TP
                                     Recall =
                                                         TP + FN

              Which metric is more important— Recall or Precision?
                 • Cases where cost of False Negative > False Positive:

                    ✶ Can be riskier and more unfavourable: As nobody expected Board exams and there was no Prediction also but
                   still Board Exams were conducted.
                    ✶ Can be costly and harmful: When no rain is predicted and the farmers did not start the harvest as they
                   thought that since no rain is predicted so their wheat crop which is ready to harvest can be done in a day or
                   two. The model that was supposed to predict the rain but did not do so spoiled the whole crop.


                        316   Touchpad Artificial Intelligence-X
   313   314   315   316   317   318   319   320   321   322   323