Otwarty dostęp

Prediction modeling using deep learning for the classification of grape-type dried fruits

Md Nurul Raihen
Md Nurul Raihen
 oraz 
Sultana Akter
Sultana Akter
   | 31 paź 2023
International Journal of Mathematics and Computer in Engineering's Cover Image
O artykule
Poprzedni artykuł
Następny artykuł
Zacytuj
Article Category: Original Study
Data publikacji: 31 paź 2023
Zakres stron: 1 - 12
Otrzymano: 20 sie 2023
Przyjęty: 19 wrz 2023
DOI: https://doi.org/10.2478/ijmce-2024-0001
Słowa kluczowe
© 2024 Md Nurul Raihen et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Fig. 1

Box plots of raisin varieties on the features.
Box plots of raisin varieties on the features.

Fig. 2

Decision Making Methodology.
Decision Making Methodology.

Fig. 3

Correlation matrix for the Raisin dataset.
Correlation matrix for the Raisin dataset.

Fig. 4

The proposed steps for creating the machine learning classification model.
The proposed steps for creating the machine learning classification model.

Fig. 5

10-fold cross Validation of the Training Dataset.
10-fold cross Validation of the Training Dataset.

Fig. 6

Accuracy of the models in predicting the Raisin dataset using a 10-fold and 5-fold cv.
Accuracy of the models in predicting the Raisin dataset using a 10-fold and 5-fold cv.

Confusion matrix.

Predictive Positive Predictive Negative
Actual Positive True Positive (TP) False Negative (FN)
Actual Negative False Positive (FP) True Negative (TN)

Outcomes from prior studies performances.

Authors Collection of Datasets (Samples) Methods that are applied Model Performance
Cinar et al., 2020 Used the system’s camera to snap images of the raisins. LR, MLP and SVM 86.44%
Dirik et al., 2023 900 raising grains and 2 classes KNN, RT and PSO-ANN 100%
Karimi et al., 2017 1400 images of raisins and top 50 features ANN and SVM 92.71%
Mollazade et al., 2012 Four different types of raisins’ color pictures were analyzed, yielding 36 colors and 8 form attributes ANN, SVM, DT and NB 96.33%
Omid et al., 2010 Obtaining raisins’ size and color characteristics using classification techniques Image Processing Technique 96%
Tarakci et al., 2021 Database for machine learning at UCI different KNN and WKNN 91.70%
Yavuj et al., 2023 Raisin dataset 2022 from UCI machine Learning repository RF and DT 85.44%
Yu et al., 2011 Separated the data into four groups based on appearance, texture, and wrinkling SVM 95%
Koklu et al. 2021 Images of pumpkin seeds LR, MLP, SVM, RF, KNN 86.64%

j.ijmce-2024-0001.tab.007

  Start
Require: n ≥ 0 data set collection
Ensure: Training of machine learning classifier
  while Evaluation of classifier with dataset? do
    if Yes then
      Prediction for raisin grains based on machine learning
    else if No then
      Parameter tuning
    end if
  end while

The metric used in evaluating the performance of machine and deep learning classifiers.

Performance Measure Name Formula
Correct Classification Rate CCR=TP+TNTP+FP+FN+TN CCR=\frac{TP+TN}{TP+FP+FN+TN}
Precision PPV=TPTP+FP PPV=\frac{TP}{TP+FP}
Recall =TPTP+FN =\frac{TP}{TP+FN}
F1-score F1=2TP2TP+FP+FN {{F}_{1}}=\frac{2TP}{2TP+FP+FN}
True Positive Rate TPR=TPTP+FN TPR=\frac{TP}{TP+FN}
False Positive Rate FPR=FPTN+FP FPR=\frac{FP}{TN+FP}
Specificity TPR=TNTN+FP TPR=\frac{TN}{TN+FP}
Negative Predictive Value NPV=TNTN+FN NPV=\frac{TN}{TN+FN}

Names and descriptions of dataset attributes.

Attribute Name Attribute Description
Area Determine how many pixels are contained within the raisin and return that value.
MajorAxisLength The maximum length of a line that can be drawn on a raisin.
MinorAxisLength The minimum length of a line that can be drawn on a raisin.
Eccentricity An ellipse, which shares the same moments as a raisin, can be described by its degree of roundness.
ConvexArea Provide the size in pixels of the smallest convex shell that contains the raisin region.
Extent Provide the fraction of the bounding box’s pixels that are within the raisin region.
Perimeter Its circumference can be determined by measuring the distance between the pixels that make up the raisin’s circumference.
Class Kecimen and Besni raisin.

Model performance with a 10-fold cv.

Model Name Accuracy (%) Precision (%) Recall (%) F1-score (%) AUC-ROC score (%)
Support Vector Machine 87.78% 90.37% 85.92% 88.00% 87.88%
Decision Tree 86.3% 91.85% 82.67% 87.02% 86.75%
Logistic Regression 87.41% 88.15% 86.86% 87.5% 87.42%
Naive Bayes 84.81% 90.37% 81.33% 85.62% 85.25%
K-nearest Neighbours 87.04% 90.37% 84.72% 87.46% 87.20%
Random Forest 86.30% 88.15% 85.00% 86.55% 86.35%
AdaBoost 90.30% 87.41% 85.51% 86.45% 86.31%
XgBoost 83.70% 85.19% 82.73% 83.95% 83.73%
LightGBM 98.40% 97.41% 89.19% 93.10% 92.57%
Convolution Neural Net. 81.35% 83.15% 79.08% 81.23% -
Radial Basis Function Net. 83.41% 84.57% 81.73% 83.78% -
Recurrence Neural Net. 73.94% 74.52% 72.50% 73.11% -
Artificial Neural Net. 65.00% 67.01% 64.86% 65.35% -
Deep Neural Net. 69.00% 70.56% 67.19% 68.89% -

Model performance with a 5-fold cv.

Model Name Accuracy (%) Precision (%) Recall (%) F1-score (%) AUC-ROC score (%)
Support Vector Machine 88.52% 91.85% 86.11% 88.88% 88.69%
Decision Tree 85.93% 88.89% 83.92% 86.33% 86.05%
Logistic Regression 86.67% 88.89% 88.37% 86.95% 86.74%
Naive Bayes 85.93% 91.11% 82.55% 86.61% 86.32%
K-nearest Neighbours 86.30% 91.11% 83.11% 86.92% 86.64%
Random Forest 85.56% 89.63% 82.88% 86.12% 85.79%
AdaBoost 89.15% 92.59% 85.03% 82.41% 88.45%
XgBoost 83.33% 87.41% 80.82% 83.98% 83.56%
LightGBM 96.31% 97.11% 88.21% 93.35% 91.83%
Convilution Neural Net. 78.51% 82.22% 75.54% 78.91% -
Radial Basis Function Net. 81.11% 83.31% 79.02% 82.37% -
Recurrence Neural Net. 71.29% 74.10% 68.53% 67.40% -
Artificial Neural Net. 64.81% 65.34% 62.13% 63.92% -
Deep Neural Net. 68.27% 70.59% 66.19% 67.78% -
eISSN:
2956-7068
Język:
Angielski
Częstotliwość wydawania:
2 razy w roku
Dziedziny czasopisma:
Computer Sciences, other, Engineering, Introductions and Overviews, Mathematics, General Mathematics, Physics
Kanał RSS czasopisma