METODE ENSEMBLE LEARNING TEKNIK WEIGHTED AVERAGE DENGAN PEMBELAJARAN SQUEEZENET PADA ARSITEKTUR MOBILENETV2 DAN EFFICIENTNET-B0 DALAM KLASIFIKASI PENYAKIT MATA

AGATHA, LUCY CHANIA and Suprihatin, Bambang and Amran, Ali (2025) METODE ENSEMBLE LEARNING TEKNIK WEIGHTED AVERAGE DENGAN PEMBELAJARAN SQUEEZENET PADA ARSITEKTUR MOBILENETV2 DAN EFFICIENTNET-B0 DALAM KLASIFIKASI PENYAKIT MATA. Undergraduate thesis, Sriwijaya University.

Preview	Image RAMA_44201_08011282126040_cover.jpeg - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (202kB) | Preview
	Text RAMA_44201_08011282126040.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (24MB) | Request a copy
	Text RAMA_44201_08011282126040_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (12MB) | Request a copy
	Text RAMA_44201_08011282126040_0013126603_0026017102_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (5MB)
	Text RAMA_44201_08011282126040_0013126603_0026017102_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (5MB) | Request a copy
	Text RAMA_44201_08011282126040_0013126603_0026017102_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (950kB) | Request a copy
	Text RAMA_44201_08011282126040_0013126603_0026017102_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (8MB) | Request a copy
	Text RAMA_44201_08011282126040_0013126603_0026017102_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (287kB) | Request a copy
	Text RAMA_44201_08011282126040_0013126603_0026017102_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (2MB) | Request a copy

Abstract

Eye diseases such as cataracts, diabetic retinopathy (DR), glaucoma are conditions that cause visual impairment to blindness. Automatic early detection can be done by classifying eye images, one of which uses Deep Learning methods. Effective deep learning approaches in eye image classification include Convolutional Neural Network (CNN). Various CNN architectures such as MobileNetV2 and EfficientNet-B0, have been developed to improve classification performance. MobileNetV2 is known as a lightweight architecture with high computational efficiency through Depthwise Separable Convolution, but has limitations in capturing the relationship between channels. EfficientNet-B0 uses Compound Scaling and Mobile Inverted Bottleneck Convolution (MBConv) to achieve computational efficiency, but the relatively small number of parameters makes EfficientNet-B0 less optimal in handling complex datasets. The performance of MobileNetV2 and EfficientNet-B0 can be improved by combining the classification results of both architectures using ensemble learning methods. This research applies ensemble learning weighted average technique through SqueezeNet learning. SqueezeNet was used for learning because it is a simple architecture that relies on fire modules, making the model faster to train and reducing the risk of overfitting in learning the weights. The average result of ensemble learning performance obtained 95% accuracy indicates the model is very good at predicting eye diseases correctly. Sensitivity of 95% indicates the model is sensitive to the normal class. Specificity of 98% indicates accuracy in predicting eye disease classes. F1-score 95% indicates the model is balanced in distinguishing each class. Cohen's Kappa 93% shows the consistency of the prediction with the actual class. These results provide an improvement over the single classification results with an accuracy of 14%, specificity of 16%, sensitivity of 5%, F1-score of 15%, and Cohen's Kappa of 20%. The performance evaluation results per class show very good performance, but it is still quite low in the DR class, which is still below 90%. The results prove that the proposed ensemble learning method is effective in the classification of eye diseases and is able to improve the performance results of a single classification.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	MobileNetV2, EfficientNet-B0, SqueezeNet, Weighted Average
Subjects:	Q Science > QA Mathematics > QA8.9-QA10.3 Computer science. Artificial intelligence. Computational complexity. Data structures (Computer scienc. Mathematical Logic and Formal Languages
Divisions:	08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User:	Lucy Chania Agatha
Date Deposited:	26 May 2025 06:17
Last Modified:	26 May 2025 06:17
URI:	http://repository.unsri.ac.id/id/eprint/174146

Actions (login required)

View Item