Claim Missing Document
Check
Articles

Found 2 Documents
Search

Klasifikasi Persediaan Stok Darah Menggunakan Algoritma K-NN, Decision Tree, dan JST Backpropagation Rijal Fauzan, Yulis; Fajarendra, Yusril Iza; Ridha , M Noor Tasiur; 'Uyun, Shofwatul
JUPITER (Jurnal Penelitian Ilmu dan Teknologi Komputer) Vol 16 No 2 (2024): Jurnal Penelitian Ilmu dan Teknologi Komputer (JUPITER)
Publisher : Teknik Komputer Politeknik Negeri Sriwijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.5281/zenodo.13755935

Abstract

The demand for blood is critical for various purposes, such as surgeries, transplants, cancer treatments, dialysis, and disaster victims. The availability of blood at the Blood Transfusion Unit (UTD) of the Indonesian Red Cross (PMI) is crucial, as a shortage of stock can endanger patients' lives. Therefore, this study aims to evaluate the condition of blood stock to determine whether it is safe or insufficient. This research focuses on comparing blood stock classification at PMI Kota Yogyakarta using three algorithms: K-Nearest Neighbor, Decision Tree, and Artificial Neural Network (Backpropagation). The study objects consist of 216 blood stock data point. Testing is conducted using the K-Fold Cross Validation method with a k value of 8 on 216 data points. The research results show that the K-Nearest Neighbors (KNN) algorithm achieves an Accuracy of 85.18%, Recall of 85.03%, Precision of 89.25%, F1-Score of 87.09%, and Specificity of 84.39%. The Decision Tree algorithm achieves an Accuracy of 84.72%, Recall of 88.18%, Precision of 86.15%, F1-Score of 87.15%, and Specificity of 78.08%. The Artificial Neural Network (Backpropagation) algorithm shows the best performance with an Accuracy of 93.05%, Recall of 96.06%, Precision of 92.42%, F1-Score of 94.20%, and Specificity of 89.35%. Thus, it can be concluded that the Artificial Neural Network (Backpropagation) algorithm outperforms the other algorithms in classifying blood stock availability.  Keywords—PMI, Blood, Classification, K-Nearest Neighbor, Decision Tree, Backpropagation
Penerapan Metode Ensemble Learning dalam Klasifikasi Risiko Abrasi Menggunakan Citra Satelit Google Earth Engine Fajarendra, Yusril Iza; 'Uyun, Shofwatul
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 13 No 1: Februari 2026
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2026131

Abstract

Abrasi menjadi masalah utama yang mempengaruhi ekosistem dan pemukiman di wilayah pesisir dengan dampak kemunduran garis pantai yang mengancam bangunan dan ekosistem yang ada didalamnya. Permasalahan utama terletak pada pemantauan, analisis dan klasifikasi risiko abrasi secara akurat menggunakan citra satelit. Data citra dengan resolusi tinggi membutuhkan komputasi yang efisien. Keterbatasan akan jumlah data adalah faktor utama yang menyebabkan model overfitting sehingga dilakukan penerapan teknik augmentasi data untuk menghasilkan sampel data sintetis dan meningkatkan kemampuan generalisasi model. Penelitian ini menggunakan data citra satelit Sentinel-2 yang diambil dari Google Earth Engine dan Google Colab untuk pemotongan dan serta dilakukan pelabelan data, dengan tiga kelas tingkatan abrasi: rendah, sedang, dan tinggi yang memiliki karakteristik citra yang berbeda. Langkah awal adalah evaluasi lima arsitektur CNN (Xception, InceptionV3, MobileNet, DenseNet, dan VGG16) melalui Transfer Learning dan K-Fold Cross-Validation. Hasilnya menunjukkan kinerja yang bervariasi, mengindikasikan tidak ada model tunggal yang optimal untuk dataset abrasi yang kompleks. Menanggapi keterbatasan ini, pendekatan (Boosting) Ensemble Learning diterapkan untuk membangun model yang lebih stabil dan general, dengan tujuan menggabungkan kekuatan prediksi berbagai arsitektur. Meskipun DenseNet menjadi model tunggal terbaik dengan akurasi 95,13%, penerapan Boosting Ensemble berhasil meningkatkan performa signifikan hingga 96,45%. Hasil ini membuktikan sinergi model memberikan solusi yang lebih unggul dan andal dibandingkan model tunggal.   Abstract Abrasion is a major problem affecting ecosystems and settlements in coastal areas, with the impact of shoreline retreat threatening buildings and the ecosystem within them. The main problem lies in the accurate monitoring, analysis, and classification of abrasion risks using satellite imagery. High-resolution imagery data requires efficient computing. Limitations in the amount of data are the main factor causing model overfitting, so data augmentation techniques are applied to generate synthetic data samples and improve model generalization capabilities. This study uses Sentinel-2 satellite imagery data taken from Google Earth Engine and Google Colab for data slicing and labeling, with three classes of abrasion levels: low, medium, and high, which have different image characteristics. The initial step was the evaluation of five CNN architectures (Xception, InceptionV3, MobileNet, DenseNet, and VGG16) through Transfer Learning and K-Fold Cross-Validation. The results showed varying performance, indicating that there is no single optimal model for complex abrasion datasets. In response to this limitation, an Ensemble Learning (Boosting) approach was applied to build a more stable and general model, with the aim of combining the predictive power of various architectures. Although DenseNet was the best single model with 95.13% accuracy, applying Ensemble Boosting significantly improved performance to 96.45%. This result demonstrates that model synergy provides a superior and more reliable solution than a single model.