Claim Missing Document
Check
Articles

Found 12 Documents
Search

Comparison of Modern NLP with Classical Machine Learning Algorithms in Evaluating Food Security Programs Sinaga, Anita Sindar; Sijabat, Dameria Esterlina; Saputri, Bella; Aulia, Nadia
Jurnal Informatika dan Rekayasa Perangkat Lunak Vol. 6 No. 4 (2025): Volume 6 Number 4 Desember 2025
Publisher : Universitas Teknokrat Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33365/jatika.v6i4.1395

Abstract

The success of food security programs faces various challenges. Most of the available data is in the form of unstructured text reports, news, and policy documents. The BERT (Bidirectional Encoder Representations from Transformers) model allows the system to read reports and news by considering the relationship between words in sentences. Compared to Support Vector Machines (SVMs) that rely on numerical data. The dataset is expanded to improve the generalization of the IndoBERT Classifier. There are 6 commodity data and 3 labels used in IndoBERT Modeling, represented by a 768-dimensional feature vector resulting in Accuracy 0.8333 (83.33%) indicating 5 correct predictions, with one misclassification. Tuned Min-Max on Support Vector Machines (SVM) is used in each dimension to find the optimal hyperplane contributing. The feature matrix x with size (39,10) and the target variable y with size (39) show Accuracy 0.92 (92.0%) that the data division process maintains the class proportion consistently. SVM performed better than IndoBERT. Classification evaluation of the models showed IndoBERT with Accuracy 83% and SVM Sccuracy 87%.
Synthetic Data Pattern Simulation of Patient Care Journey Using K-Means Clustering Sitio, Arjon Samuel; Parlindungan, Richard; Sinaga, Anita Sindar
Jurnal Informatika dan Rekayasa Perangkat Lunak Vol. 6 No. 4 (2025): Volume 6 Number 4 Desember 2025
Publisher : Universitas Teknokrat Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33365/jatika.v6i4.1498

Abstract

Heterogeneous synthetic data is artificial data that can include many types of features (demographics, examinations, therapies). Complex patients (many procedures & medications) but fast service process and low complications. All patients are divided into 4 clusters, patient segmentation includes cluster 1 including mild patients, Cluster 2 including complex patients, Cluster 3 including high costs, Cluster 4 including high readmission risk. The highest silhouette score is 0.2187, which is obtained when the number of clusters (k) is 2. Based on previous calculations, the Davies-Bouldin Index result for the current clustering solution is 2.33. The Calinski-Harabasz index for the clustering solution with k=4 is 367.72. Clustering results are simply groups, without labels. Further analysis is needed to assign clinical meaning to each cluster.