Text

ENSEMBLE LEARNING MENGGUNAKAN WEIGHTED VOTING DENGAN PEMBELAJARAN PYRAMIDNET PADA HASIL INCEPTIONV3, MOBILENETV2, DAN VISION TRANSFORMER DALAM KLASIFIKASI CITRA KANKER PAYUDARA

---

Penilaian

---

Breast cancer is one of the leading causes of death in women. Early detection can be done by classifying ultrasound images (USG) of breast cancer into 3 levels of severity namely normal, benign, and malignant using Deep Learning (DL). DL methods include Convolutional Neural Network (CNN) and transformers. CNN architecture development includes InceptionV3 and MobileNetV2. InceptionV3 recognizes image patterns with various scales using factorized convolutions, but requires large memory and long computation time. MobileNetV2 is more efficient in memory and computation time. However, the bottleneck layer may cause the loss of some important details. CNN uses convolution operations with kernel and stride sizes, so it is limited in capturing global relationships. The transformer method used is Vision Transformer (ViT). ViT captures global relationships by dividing the image into small patches, which are then processed in parallel using self-attention. However, ViT gives higher weight to frequently occurring patterns, making it less optimal for unbalanced data. The weakness of a single architecture can be overcome with ensemble learning. This study applies ensemble learning using weighted voting through PyramidNet. PyramidNet helps the model learn the weights on each data and avoid excessive training. The average ensemble learning performance results obtained 93% accuracy indicating almost completely correct class predictions. Sensitivity 93% shows the model is excellent at classifying data of a certain class. Specificity 93% shows the model is excellent at classifying data that is not a certain class. F1-score of 93% shows the model is balanced in distinguishing each class. Cohen's Kappa of 89% indicates high agreement with the true class. This method improved the performance over a single classifier, with accuracy increasing by14%, sensitivity of 16.66%, F1-score of 17.33%, and Cohen's Kappa of 24.33%. The class performance exceeded 90% across all evaluation metrics. However, the Cohen’s Kappa for each class remained below 90%. The model best predicted the normal class, followed by the malignant, and benign classes. The results prove that the proposed ensemble learning method is effective in the classification of breast cancer ultrasound images.

---

Availability

Inventory Code	Barcode	Call Number	Location	Status
2507002023	T169919	T1699192025	Central Library (Reference)	Available but not for loan - Not for Loan

Detail Information

Series Title

-

Call Number

T1699192025

Publisher

: Prodi Ilmu Matematika, Fakultas Matematika Dan Ilmu Pengetahuan Alam Universitas Sriwijaya., 2025

Collation

xi, 80 hlm.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

510.07

Content Type

-

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Matematika

Specific Detail Info

-

Statement of Responsibility

EM

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment