Intense sun exposure is a major risk factor for the development of melanoma, an abnormal proliferation of skin cells. Yet, this more prevalent type of skin cancer can also develop in less-exposed areas, such as those that are shaded. Melanoma is the sixth most common type of skin cancer. In recent years, computer-based methods for imaging and analyzing biological systems have made considerable strides. This work investigates the use of advanced machine learning methods, specifically ensemble models with Auto Correlogram Methods, Binary Pyramid Pattern Filter, and Color Layout Filter, to enhance the detection accuracy of Melanoma skin cancer. These results suggest that the Color Layout Filter model of the Attribute Selection Classifier provides the best overall performance. Statistics for ROC, PRC, Kappa, F-Measure, and Matthews Correlation Coefficient were as follows: 90.96% accuracy, 0.91 precision, 0.91 recall, 0.95 ROC, 0.87 PRC, 0.87 Kappa, 0.91 F-Measure, and 0.82 Matthews Correlation Coefficient. In addition, its margins of error are the smallest. The research found that the Attribute Selection Classifier performed well when used in conjunction with the Color Layout Filter to improve image quality.
Keywords: Color layout filter; Auto color correlogram filter; Attribute selection classifier; Binary pattern pyramid filter; Bagging
That much is certain: the skin is the body's largest organ. The body's critical organs are protected from the elements. The skin shields us from the sun's damaging rays, keeping our core body temperature consistent. [[
Significance of the study, the authors of this work have used the Auto Correlogram Methods, the Binary Pyramid Patter Filter, and the Ensemble Model to increase the accuracy of their detection of Melanoma skin cancer. The Color Layout Filter model of the Attribute Selection Classifier provides the best overall performance, with statistics for ROC, PRC, Kappa, F-Measure, and Matthews Correlation Coefficient showing 90.96% accuracy, 0.91 precision, 0.91 recall, 0.95 ROC, 0.87 PRC, 0.87 Kappa, 0.91 F-Measure, and 0.82 Matthews Correlation Coefficient. The research also found that the Attribute Selection Classifier performed well when used in conjunction with the Color Layout Filter to improve image quality.
Various AI methods, such as the construction of multi-layered structures of input and output training data [[
One of the most important areas where ANNs have been put to use is in the field of diagnosing diseases [[
This proposed system was implemented by ISIC 2018. This work considers around 10,000 images. They are specified in below Table 1.
Table 1 Meta data of ISIC (International Skin Imaging Collaboration) dataset
S. No. Name of the class Description 1 nv Melanocytic nevi 2 mel Melanoma 3 bkl Benign keratosis lesions 4 bcc Basal cell carcinoma 5 akiec Actinic keratoses 6 vasc Vascular 7 df Dermatofibroma
The following techniques are applied in this research work.
- Image Acquisition
- Image preprocessing
- Apply Auto Color Correlogram Filter, Binary Patterns Pyramid Filter, and Color Layout Filter by producing 34 attributes.
- Relate for machine learning algorithms
- Attribute selection is used to minimize the dimensionality of both training and test data before they are handed on to a classifier.
- Bagging-Class is used to reduce variation in a classifier by bagging.
- To get an optimal solution.
To produce a final result, these techniques have been implemented in one of the top and open supply programs, Weka3.9.5. This observation makes use of handiest 10% of the complete dataset and makes use of tenfold go validation for all categories.
Table 2 summarizes the findings of this research study. This experimentation is recognized by relating numerous [[
Table 2 Performance of classifiers on dataset by auto color correlogram filter
S. No. Ensemble classifier Accuracy (%) Precision Recall ROC PRC 1 Bagging with auto color correlogram filter 83.88 0.85 0.84 0.92 0.90 2 Attribute selected classifier with auto color correlogram filter 82.65 0.82 0.83 0.92 0.72 3 Bagging with binary patterns pyramid filter 85.06 0.84 0.85 0.90 0.88 4 Attribute selected classifier with binary patterns pyramid filter 89.97 0.90 0.90 0.91 0.91 5 Bagging with color layout filter 85.77 0.85 0.86 0.87 0.88 6 Attribute selected classifier with color layout filter 90.96 0.91 0.91 0.95 0.87
Graph: Fig. 1Methodology proposal
Graph: Fig. 2Representation of dataset in Weka.3.9.5
Graph: Fig. 3Sample dataset
Table 2 displays the results of using a few different image enhancing algorithms with a few different classifiers. It has been found that the Attribute Selected Classification algorithm with the implementation of the Auto Color Correlogram Filter achieves an accuracy level of 82.65%, the Bagging algorithm with the use of the Binary Patterns Pyramid Filter of image feature extraction achieves an accuracy level of 85.06%, and the Bagging algorithm with the use of the Auto Color Correlogram Filter of meta category classification achieves an accuracy level of 83.88%.
Table 2 displays the accuracy scores achieved by various classifiers using various image enhancing strategies. Precision levels for the Auto Color Correlogram Filter-based Bagging of meta-category classification algorithms are 0.85 and 0.82, respectively; the Binary Patterns Pyramid Filter-based Bagging of image feature extraction algorithms is 0.84, and the Attribute Selected Classification algorithms is 0.82.
Table 2 shows the classifiers that were chosen alongside the corresponding picture improvement methods. The recall levels of the Attribute Selected Classification and Bagging algorithms are 0.83 and 0.84, respectively, while the recall levels of the Bagging and Binary Patterns Pyramid Filter of image feature extraction are 0.85 and 0.84, respectively.
Table 2 displays the ROC values achieved by the chosen classifiers using the chosen image enhancing methods. The ROC for the Attribute Selected Classification algorithm that uses an Auto Color Correlogram Filter is 0.92, the ROC for the Bagging algorithm that uses an image feature extraction method based on a Binary Patterns Pyramid Filter is 0.90, and the ROC for the Bagging algorithm that uses an Auto Color Correlogram Filter for meta category classification is 0.92.
Table 2 displays the PRC values produced using various image enhancement methods and several classifiers. For example, the PRC level of 0.90 is achieved by the Bagging of meta-category classification algorithms using the Auto Color Correlogram Filter, while the PRC level of 0.72 is achieved by the Attribute Selected Classification algorithms using the Auto Color Correlogram. The PRC level of 0.88 is achieved by the Bagging of image feature extraction algorithms using the Binary Patterns Pyramid Filter.
Figure 4 depicts the results of using the recommended classifiers in conjunction with various picture feature extraction methods in terms of accuracy. This chart compares the accuracy of different classifier ensembles employing different picture filters. The Attribute Selection Classifier algorithm employing the Auto Color Correlogram Filter generates the least accurate value of 82.65 percent. Attribute Selection Classifier with Color Layout filter has the greatest accuracy of 90.96%. The accuracy scores range from 83.88 percent to 89.97 percent for the Bagging with Auto Color Correlogram Filter, Bagging with Binary Patterns Pyramid Filter, Bagging method with Color Layout Filter, and Attribute Selection Classifieralgorithm with Binary Patterns Pyramid Filter.
Graph: Fig. 4Accuracy performance of classifiers
Figure 5 displays the accuracy values derived from the chosen classifiers using the chosen picture feature extraction methods. All the groups of classifiers using different image filters are compared in terms of precision here using a graphical representation. The Attribute Selection Classifier algorithm employing the Auto Color Correlogram Filter yields an accuracy of 0.82 at its lowest setting. Attribute Selection Classifier with Color Layout Filter achieves the greatest precision of 0.91. Classifiers like Bagging with Auto Color Correlogram Filter, Bagging with Binary Patters Pyramid Filter, Bagging method utilizing Color Layout Filter, and Attribute Selected Classifier algorithm using Binary Patters PyramidFilter all have precision levels between 0.85 and 0.90.
Graph: Fig. 5Precision performance of classifiers
The aforementioned chart depicts the classifiers' recall performances after being chosen. Here, we see how different classes of classifiers using different image filters compare with respect to recall rates. With the Auto Color Correlogram Filter, the Attribute Selection Classifier method generates a recall value of 0.83, which is the lowest possible. With an Attribute Selection Classifier using a Color Layout Filter, we achieve a recall of 0.91. Recall values for the remaining classifiers range from 0.84 on the recall scale to 0.90 on the recall scale, and they include The utilization of various bagging algorithms, namely Bagging with Auto Colour Correlogram Filter, Bagging with Binary Patterns Pyramid Filter, Bagging method using Colour Layout Filter, and Attribute Selected Classifier algorithm using Binary Patterns Pyramid Filter, has been investigated. (see Fig. 6).
Graph: Fig. 6Recall performance of classifiers
Figure 7 displays the ROC values obtained by various classifiers using various image feature extraction strategies. This diagram depicts the comparison of ROC values across all categories of classifiers employing different image filters. With the Color Layout Filter, the Bagging method generates a ROC value of 0.87, which is the lowest possible. The highest ROC value is 0.95, which is having an Attribute Selected Classifier by implementing a Colour Layout Filter.
Graph: Fig. 7ROC performance of classifiers
The ROC values for the other classifiers range from 0.90 to 0.92, and they include the Bagging using the Binary Patters Pyramid Filter and the Attribute Selection Classifier method using the Binary Patterns Pyramid Filter. In this case, the ROC value for using the Auto Color Correlogram Filter for either bagging or crediting the selected classifier is the same.
Figure 8 displays the PRC values derived using the aforementioned classifiers and feature extraction methods for images. This chart shows the comparison of PRC scores across all categories of classifiers employing different image filters. Attribute Selection Classifier with Auto Color Correlogram Filter generates the lowest PRC value of 0.72. Attribute Selection Classifier using Binary Patterns Pyramid Filter yields the greatest PRC value of 0.91.
Graph: Fig. 8PRC performance of classifiers
PRC values range from 0.87 to 0.90 for the Attribute Chosen Classifier with Colour Layout Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with Colour Layout Filter, and Bagging with Auto Color Correlogram Filter. The PRC value of 0.88 is shared by two models: the Bagging algorithm using the Color Layout Filter and the Bagging method using the Binary Patterns Pyramid Filter.
Table 3 displays the kappa values obtained from the various classifiers using the chosen image enhancement methods. Auto Color Correlogram Filter is used in the Bagging of meta category classification algorithms, yielding a 0.70 kappa statistic value; its implementation in the Attribute Selected Classification algorithms yields a 0.66 kappa statistic value; the Binary Patterns Pyramid Filter is used in the Bagging of image feature extraction algorithms, yielding a 0.71 kappa statistic value; and the Attribute Selected Classification algorithms yield a 0.66 kappa statistic value.
Table 3 Kappa, F1 Score, MCC, and performance of classifiers
S. No. Ensemble classifier Kappa statistic F-measure MCC Time taken to build model 1 Bagging with auto color correlogram filter 0.70 0.84 0.73 2.80 2 Attribute selected classifier with auto color correlogram filter 0.66 0.80 0.79 4.19 3 Bagging with binary patterns pyramid filter 0.71 0.84 0.74 1.00 4 Attribute selected classifier with binary patterns pyramid filter 0.79 0.90 0.71 17.95 5 Bagging with color layout filter 0.72 0.85 0.74 27.17 6 Attribute selected classifier with color layout filter 0.80 0.91 0.82 16.97
Table 3 displays the F-Measure values obtained from the various classifiers using the aforementioned image enhancing methods. The Attribute Selected Classification algorithm by implementing Auto Color Correlogram has 0.80F-Measure value, while the Bagging algorithm for the meta category by using the Binary Patterns Pyramid Filter of image features extraction achieves 0.84F-Measure value.
Table 3 displays the MCC values obtained using various image enhancing approaches and several classifiers. The MCC for the Attribute Selected Classification algorithm that implements the Auto Color Correlogram Filter is 0.79, and the MCC for the Bagging algorithm that uses the Binary Patterns Pyramid Filter for image feature extraction is 0.74.
Table 3 displays the time it took to construct their models using the chosen classifiers and the chosen picture enhancing methods. It took 2.80 s for the Bagging of meta category classification algorithm to build its model using Auto Color Correlogram Filter, and it took 4.19 s for the Attribute Selected Classification algorithm to build its model using the Binary Patterns Pyramid Filter image feature extraction technique. In that time, it has accumulated 1 s of data and built a model. By using the Binary Patterns Pyramid Filter method of extracting features from images, the Attribute Selection Classification algorithm can reliably classify images based on their attributes. Building a model with the Color Layout Filter of Image enhancement approach Filter took 27.17 s with the Bagging classifier, and 16.97 s with the Attribute Selection Classification algorithm.
Figure 9 displays the Kappa values attained by various classifiers employing various picture feature extraction methods. Kappa values are evaluated for every classifier group using each image filter, and the results are plotted here. Attribute Selection Classifier with Auto Color Correlogram Filter yields the lowest kappa value of 0.66. The Attribute Selected Classifier with the Color Layout Filter yielded the highest kappa statistic value (0.80).
Graph: Fig. 9Kappa performance of classifiers
There is a wide range of kappa values between 0.70 and 0.79 for the four different bagging methods (Bagging with Auto Color Correlogram Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with Color Layout Filter, and Attribute Selection Classifier with Binary Patterns Pyramid Filter).
The F-Measure values obtained from the nominated classifiers with selected image feature extraction techniques are shown in Fig. 10. This graph characterizes the contrast of F-Measure values for all the groups of the classifiers with various image filters. The least F-Measure value is 0.80, which is created by the Attribute Selected Classifier with Auto Color Correlogram Filter. The highest F-Measure value is 0.91, which is having an Attribute Selected Classifier by implementing Color Layout Filter.
Graph: Fig. 10F-Measure performance of classifiers
The F-Measure values of the Bagging with Auto Color Correlogram Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with Color Layout Filter, and Attribute Selected Classifier with Binary Patterns Pyramid Filter range from 0.84 to 0.90. Hence, bagging using Auto Color Correlogram Filter and bagging using Binary Patterns Pyramid Filter models have the same F-Measure value, which is 0.84 of the F-Measure value.
The MCC values obtained from the particular classifiers with selected image feature extraction techniques are shown in Fig. 11. This graph represents the assessments of MCC values for all the types of classifiers with various image filters. The least MCC value is 0.71, which is produced by the Attribute Selected Classifier with Binary Pattern Pyramid Filter. The highest MCC value is 0.82, which has an Attribute Selected Classifier by implementing Color Layout Filter.
Graph: Fig. 11MCC performance of classifiers
The Bagging with Auto Color Correlogram Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with Color Layout Filter, and Attribute Selected Classifier with Auto Color Correlogram have MCC values from 0.73 of the MCC value to 0.79 of the MCC value. As a result, bagging with the Color Layout Filter model and bagging with the Binary Patterns Pyramid Filter model have the same MCC value, which is 0.74 of the MCC value.
The time consumption to build their models acquired from the particular classifiers with selected image feature extraction techniques is shown in Fig. 12. This graph exemplifies the contrasts in time consumption for building models for all the categories of the classifiers with various image filters. The least time-consuming way to build a model is 1 s, which is produced by the Attribute Selected Classifier with Auto Color Correlogram Filter. The highest time consumption is 27.17 s to build a model which has Bagging by implementing Color Layout Filter.
Graph: Fig. 12Time- consumption performance of classifiers
Bagging with Auto Color Correlogram Filter, Attribute Selected Classifier with Color Layout Filter, and Attribute Selected Classifier with Binary Patterns Pyramid Filter all take between 2.80 and 17.95 s.
Table 4 displays the Mean Absolute Errors achieved by several classifiers using various image improvement strategies. Mean absolute error values for meta-category-classification-algorithm-Bagging with the Auto Color Correlogram Filter are 0.23 and 0.38, respectively; for image-feature-extraction-algorithm-Bagging with the Binary Patterns Pyramid Filter, the values are 0.24 and 0.24, respectively; and for attribute-selected-classification-algorithm-Bagging, they are 0.24 and 0.38, respectively.
Table 4 Deviation performance of classifiers
S. No. Ensemble classifier Mean absolute error Root mean squared error Relative absolute error (%) Root relative squared error (%) 1 Bagging with auto color correlogram filter 0.23 0.42 50.66 90.30 2 Attribute selected classifier with auto color correlogram filter 0.38 0.46 96.98 99.73 3 Bagging with binary patterns pyramid filter 0.24 0.40 55.05 82.96 4 Attribute selected classifier with binary patterns pyramid filter 0.25 0.37 54.51 76.18 5 Bagging with colour layout filter 0.27 0.41 62.25 86.34 6 Attribute selected classifier with colour layout filter 0.24 0.30 53.73 74.40
Table 4 displays the Root Mean Squared Error scores achieved by several classifiers using various image enhancing strategies. Using the Auto Color Correlogram Filter in a meta-category classification algorithm yields an RMS error of 0.42, while doing the same with the Attribute Selected Classification algorithm yields an RMS error of 0.46, while using the Binary Patterns Pyramid Filter in an image feature extraction technique yields an RMS error of 0.40.
Table 4 displays the relative absolute errors achieved by several classifiers employing various picture improvement strategies. It has been found that the Attribute Selected Classification algorithm, when using Auto Color Correlogram, has a relative absolute error value of 96.98%, while the Bagging algorithm, when using the Binary Patterns Pyramid Filter of the image feature extraction technique, has a relative absolute error value of 55.05%.
Table 4 displays the Root Mean Square Errors achieved by several classifiers using various picture enhancing methods. The root relative squared value is 90.30% for the Bagging of the meta category classification algorithm using the Auto Color Correlogram Filter, and it is 99.73% for the Attribute Selected Classification algorithm using the same filter. The Bagging of the image feature extraction technique using the Binary Patterns Pyramid Filter yields an even lower value of 82.96% for the root relative squared value.
The Mean Absolute Error values obtained from the selected classifiers with selected image feature extraction techniques are shown in Fig. 13. This graph represents the comparisons of MAE values for all the categories of the classifiers with various image filters. The least mean absolute error value is 0.23, which is produced by bagging with the Auto Color Correlogram Filter. The highest MAE value is 0.38, which is having an Attribute Selected Classifier by implementing the Auto Color Correlogram Filter. The Bagging with Binary Patterns Pyramid Filter, Attribute Selected Classifier with color Layout Filter, Attribute Selected Classifier with Binary Patterns Pyramid Filter, and Bagging with color Layout Filter have MAE values ranging from 0.24 of mean absolute value to 0.27 of MAE value. Hence, bagging using Binary Patterns Pyramid Filter and bagging using Attribute Selected Classifier with color Layout Filter models have the same MAE value, which is 0.24 of the MAE value.
Graph: Fig. 13MAE performance of classifiers
The Root Mean Squared Error values obtained from the selected classifiers with selected image feature extraction techniques are shown in Fig. 14. It represents the comparisons of RMSE values for all the categories of the classifiers with various image filters. The least RMSE value is 0.30, which is produced by the Attribute Selected Classifier with color Layout Filter. The highest RMSE value is 0.46, which is having an Attribute Selected Classifier by implementing Auto Color Correlogram Filter. The Attribute Selected Classifier with Binary Patterns Pyramid Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with color Layout Filter, and Bagging with Auto Color Correlogram Filter has a RMSE of 0.37 of RMSE to 0.42 of RMSE.
Graph: Fig. 14RMSE performance of classifiers
The relative absolute error values obtained from the selected classifiers with selected image feature extraction techniques are shown in Fig. 15. This graph represents the comparisons of RAE values for all the categories of the classifiers with various image filters. The least RAE value is 50.66% of the RAE value, which is produced by Bagging with Auto Color Correlogram Filter. The highest RAE value is 96.98% of the RAE value, which is having an Attribute Selected Classifier by implementing Auto Color Correlogram Filter.
Graph: Fig. 15RAE performance of classifiers
The Attribute Selected Classifier with color Layout Filter, Attribute Selected Classifier with Binary Patterns Pyramid Filter, Bagging with Binary Patterns Pyramid Filter, and Bagging with color Layout Filter have RAE values from 53.73% of RAE value to 62.25% of RAE value.
The Root Relative Squared Error values obtained from the selected classifiers with selected image feature extraction techniques are shown in Fig. 16. It represents the comparisons of RRSE values for all the categories of the classifiers with various image filters. The least RRSE value is 74.40% of RRSE, which is produced by the Attribute Selected Classifier with Color Layout Filter. The highest root relative squared error is 99.73% of the RRSE value, which is having an Attribute Selected Classifier by implementing Auto Color Correlogram Filter.
Graph: Fig. 16RRSE performance of classifiers
The Attribute Selected Classifier with Binary Patterns Pyramid Filter, Bagging with Binary Patterns Pyramid Filter, Bagging with Color Layout Filter and Bagging with Auto Color Correlogram Filter are RRSE values from 76.18% of RRSE value to 90.30% of RRSE value.
The previous research [[
The present study demonstrates that the Attribute Selected Classifier, belonging to the ensemble category, while utilizing the color Layout Filter model, yields a very effective output with a reduced number of error values. The accuracy, precision, recall, ROC, PRC, kappa statistic, F-Measure, and MCC values are 90.96%, 0.91, 0.94, 0.81, and 0.82, respectively. These values were obtained using the Attribute Selected Classifier with the implementation of the Colour Layout Filter. Skin cancer is the most commonly occurring and hazardous type of cancer in the human population. Melanoma is a type of skin cancer that has the potential to be life-threatening. Early detection greatly increases the likelihood of successful treatment and cure. The biopsy process is the established method for diagnosing melanoma. The aforementioned procedure can be characterized by its time-intensive nature and the potential for inducing discomfort. The present study presents a computer-aided detection approach for the early identification of melanoma. This study presents a diagnostic system that utilizes the Attribute Selected Classifier of Ensemble Category with the color Layout Filter model methodologies to achieve effective results. The image of the affected skin undergoes a series of preprocessing procedures before being enhanced and refined.
This work was supported by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2023R235), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University (KKU) for funding this research through the Research Group Program Under the Grant Number: (R.G.P.2/283/44).
The authors confirm contribution to the paper as follows: study conception and design: PK, BOS and GA; data collection: PJ, SKM and SM; analysis and interpretation of results: AA-R and MSA; draft manuscript preparation. All authors reviewed the results and approved the final version of the manuscript.
This research was financially supported by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2023R235), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University (KKU) for funding this research through the Research Group Program Under the Grant Number: (R.G.P.2/283/44).
The datasets used during the current study are available from the corresponding author on reasonable request.
Not applicable.
Not applicable.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
By P. Kavitha; G. Ayyappan; Prabhu Jayagopal; Sandeep Kumar Mathivanan; Saurav Mallik; Amal Al-Rasheed; Mohammed S. Alqahtani and Ben Othman Soufiene
Reported by Author; Author; Author; Author; Author; Author; Author; Author