Page 112 - IJEEE-2023-Vol19-ISSUE-1
P. 112

108 |                                                                                                      Abed, Wali, & Alaziz

                           (c) V =2.5 m/s                             algorithm is more complicated. The DT is considered a
                        Fig. 12: Continued.                           confirmation algorithm more than SVM because it does not
                                                                      deal with the dependent and independent data as linear or
  Fig. 13: Pressure distribution of various velocities, gas is        non-linear regression. While SVM should be specified for
                             used fluid.                              linear or nonlinear expressions which must be solved by
                                                                      Gaussian approximation [33], the accuracy is determined by
                                                                      the following equation:

                                                                              7889:;8< = ?@AC?@@AA??BBACB	                                    (6)	

                                                                      The following equations also determine the precision, recall,
                                                                      and F1-score for the SVM model:

                                                                                           =:>8?@?AB = ?@?A@C@	                               (7)
                                                                                                       	                                      (8)
                                                                                                                                              (9)	
                                                                                             !>8;CC = ?@?A@CB	

                                                                      Where:  D1 - @8A:>	 = 2	?	 @@DD''EFEGFFGHFH""	 ?AJJ''EE%%KKKK			

                                                                      TP is True Positive

                                                                      TN is True Negative

                                                                      FP is False Positive, and

                                                                      FN is False Negative

                                                                                                 TABLE II

                                                                      Precision, recall, and f1-score for the (SVM) and (DT)

                                                                                                  models.

                                                                      Model Precision % Recall % F1-score %

                                                                      SVM                  91.67           88                           89.8

                                                                      DT 100 97 100

                                                                           Fig. 15, shows that the optimization plots are developed
                                                                      based on trained values; the parallel coordinate of column
                                                                      interactions (position, pressure at 0.1 m/s, pressure at 1 m/s,
                                                                      and pressure at 2.5 m/s). The scatter plot indicates the
                                                                      pressure distribution of the leaked and non-leaked points are
                                                                      presented based on optimization plots and confusion matrix.

                                Ball position

Fig.  14:    Outlet Oil         between  Lleeaakks  and    Inlet Oil

           Pressure comparison                           non-leaks

           cases where oil flows in 2.5 m/s.

  B. Machine learning models results                                             (a) Parallel coordinates of optimization.

     The training phase of SVM and DT is divided into 70%                     Fig. 15: The optimization plot and Scattering
for training and 30% for testing by cross-validation, and the                      examination using the SVM model.
average accuracy is 98.8%, and 99.87%, respectively. Table
II shows the SVM and DT model's average precision, recall,
and F1-score of the present work. The DT has perfect
precision, Recall, and F1-score as compared with SVM. The
membership function of DT is more than SVM, the DT
   107   108   109   110   111   112   113   114   115   116   117