| Comparison of Different Machine Learning Algorithms Performance in Grading of Jujube Fruit |
| Paper ID : 1451-NICAME1402 |
| Authors: |
|
Sayed Javad Sajadi * Plant Production Department, Faculty of Agriculture and Natural Resources, Gonad Kavous University |
| Abstract: |
| The purpose of this study is to investigate the possibility of using impact sound processing in the time-frequency domain and to compare different feature selection and machine learning algorithms in grading jujube fruits into small, medium, large and very large groups. For this purpose, the sound of fruits falling from a height of 25 cm on a 10 mm thick metal plate was recorded. Then, short-time Fourier transform was used to perform time-frequency analysis. The extracted time-frequency features include instantaneous frequency and spectral entropy. Due to the large number of features extracted from each signal, three feature selection algorithms, ReliefF, MRMR and Chi2, were investigated. The features extracted by each feature selection algorithm were used as input to different machine learning models. In this research, decision tree models, differential analysis, simple Bayesian, support vector matching, nearest K distance, random Jenkel and neural network were evaluated. The results of the evaluation of machine learning models in the test phase showed that the SVM and KNN models trained with the features extracted by MRMR and Chi2 algorithms have similar accuracy in all the investigated criteria except the MCC criterion. The standard value of MCC of SVM and KNN models is 0.978 and 0.977, respectively, in the test phase. Based on this, it can be said that the accuracy of the SVM model was higher than the accuracy of the KNN model in jujube fruit grading with a very small difference. |
| Keywords: |
| Jujube, Machine Learning, Digital Signal Processing |
| Status : Paper Accepted (Poster Presentation) |
