Jurnal Informatika Progres

OPTIMASI KLASIFIKASI JENIS INDUSTRI KECIL MENENGAH (IKM) MENGGUNAKAN DEEP NEURAL NETWORK BERBASI SHAP ANALYSIS

Tue, 08 Sep 2026 05:29:28 +0000

Small and Medium Industries (SMEs) in Makassar City face a significant gap between high labor absorption (66.25%) and low GDP contribution (20%), often due to conventional and experience-based determination of industry types. This study implements a Deep Neural Network (DNN) model to classify four categories of SMEs, Bread and Cake, Processed Food, Vehicle Repair, and Textiles and Clothing to facilitate data-driven decisions. Using a supervised learning approach on 31,824 data samples for the 2022-2024 period, this model was developed through feedforward and backpropagation mechanisms. The results showed superior performance with overall accuracy of (93.99%) and balanced accuracy (96.70%), which signified an increase of (15.88%) compared to the Naïve Bayes baseline model. All F1-scores above (90%) indicate strong performance stability in each class. Furthermore, SHAP's analysis revealed that textual features (87.1%) were the dominant factors, followed by the type of business entity and investment value. This study confirms that DNN is effective in modeling complex non-linear interactions, providing objective tools for classification and strategic economic planning in Makassar City.

KLASIFIKASI MULTI-CLASS STATUS GIZI BALITA MENGGUNAKAN ARSITEKTUR DEEP NEURAL NETWORK

Tue, 08 Sep 2026 06:00:12 +0000

This study aims to develop a classification model for toddler nutritional status using a Deep Neural Network (DNN) with a multi-class classification approach. The research utilizes anthropometric data of toddlers aged 0-60 months obtained from UPTD Puskesmas Cendana Putih, North Luwu Regency, covering the period 2023–2025. The dataset consists of 156 records with features including age, weight, height, and Z-score indicators. Data preprocessing involves validation, normalization, and splitting into training and testing sets with a ratio of 85:15. The DNN model is constructed with multiple hidden layers (128, 64, and 32 neurons) and trained using the Adam optimizer and categorical cross-entropy loss function. The results show that the model achieves an accuracy of 91.67% on the testing data, indicating good performance in classifying nutritional status into categories such as undernutrition, normal, and obesity. Evaluation using confusion matrix and classification metrics (precision, recall, and F1-score) reveals that the model performs well on dominant classes but shows limitations in minority classes due to data imbalance. Overall, the proposed model demonstrates potential as a decision support tool to assist healthcare workers in identifying toddler nutritional status more accurately and efficiently.

ESTIMASI RENCANA ANGGARAN BIAYA PROYEK TELEKOMUNIKASI BERBASIS ALGORITMA CATBOOST DAN DATA BOQ

Tue, 08 Sep 2026 06:37:14 +0000

Estimasi Rencana Anggaran Biaya (RAB) yang akurat merupakan elemen krusial dalam keberhasilan proyek infrastruktur telekomunikasi. Penentuan harga satuan pada dokumen Bill of Quantities (BOQ) secara konvensional sering kali tidak efisien dan subjektif. Tantangan utama otomatisasi estimasi ini adalah tingginya kardinalitas fitur kategorikal berupa teks deskriptif. Penelitian ini mengusulkan penerapan algoritma CatBoost untuk memprediksi harga satuan pekerjaan berbasis data BOQ. Tahapan penelitian meliputi pembersihan data historis sebanyak 20.611 item, transformasi logaritma natural pada variabel target, serta pelatihan model dengan pembagian data 80% latih dan 20% uji. Hasil eksperimen menunjukkan CatBoost mampu menghasilkan Coefficient of Determination (R²) sebesar 97,66% dan Mean Absolute Error (MAE) sebesar Rp 13.289. Kinerja ini unggul dibandingkan algoritma pembanding XGBoost (R² 85,93% dan MAE Rp 23.614). Validasi manual mengonfirmasi rasio kesalahan prediksi hanya 0,05 dari total nilai proyek, yang membuktikan kelayakan model untuk otomatisasi RAB secara presisi.

PENERAPAN ALGORITMA HIDDEN MARKOV MODEL PADA PREDIKSI EKSPOR KOMODITI BIJI KOPI

Tue, 08 Sep 2026 07:00:24 +0000

Fluktuasi volume dan nilai ekspor komoditas biji kopi Indonesia dipengaruhi oleh faktor dinamis yang tidak dapat diamati secara langsung, seperti pergeseran kondisi pasar internasional dan dinamika iklim. Penelitian ini bertujuan untuk menerapkan algoritma Hidden Markov Model (HMM) dalam memodelkan tren tersembunyi (hidden states) serta memprediksi tren ekspor biji kopi. Dataset yang digunakan merupakan data historis bulanan ekspor biji kopi periode 2020–2024. Estimasi parameter dilakukan menggunakan algoritma Baum-Welch, sedangkan performa model dievaluasi menggunakan metrik Mean Absolute Error (MAE) dan Root Mean Square Error (RMSE) serta dibandingkan dengan metode Seasonal Naïve (S-Naïve). Hasil penelitian menunjukkan bahwa HMM mampu mengidentifikasi tiga state utama (Naik, Stabil, dan Turun) dengan nilai MAE sebesar 20,81 dan RMSE sebesar 28,45. Performa HMM melampaui metode S-Naïve yang memiliki MAE sebesar 35,12. Dengan demikian, pendekatan HMM terbukti adaptif dalam menangkap volatilitas dan transisi kondisi pasar pada komoditas biji kopi.

KLASIFIKASI RISIKO INFEKSI SALURAN PERNAPASAN AKUT MENGGUNAKAN ENSEMBLE SOFT VOTING BERBASIS REKAM MEDIS

Tue, 08 Sep 2026 07:25:43 +0000

This study proposes a risk classification model for Acute Respiratory Infections using a Soft Voting-based Ensemble Learning approach, which integrates prediction probabilities from the Random Forest and Extreme Gradient Boosting algorithms. The research utilized 456 patient medical record data from RSUD Latemmamala spanning January 2020 to December 2025. Comparative evaluation results show that the Random Forest model achieved an accuracy of 94.57%, Extreme Gradient Boosting reached 95.65%, and the Soft Voting Ensemble model delivered the best performance with an accuracy of 96.74%, precision of 96.97%, recall of 96.88%, and an F1-Score of 96.77%. Furthermore, the Soft Voting Ensemble model successfully achieved a perfect recall score for the Severe Acute Respiratory Infection category, ensuring that no high-risk patients went undetected. In conclusion, the Soft Voting Ensemble model serves as a reliable decision-support tool to assist medical professionals in triaging Acute Respiratory Infection patients quickly, objectively, and accurately.

SISTEM INFORMASI MANAJEMEN INVENTARIS BERBASIS WEB MENGGUNAKAN METODE SCRUM DAN TEKNOLOGI QR CODE

Pranata Eka Pramudya, Ummul Hairah, Indah Fitri Astuti — Tue, 08 Sep 2026 14:26:26 +0000

Manual inventory management often causes recording errors, delayed updates, and monitoring difficulties due to uncentralized data. This study aims to develop a web-based inventory management information system integrated with QR Code technology to centralize data tracking. System development was carried out using the Scrum methodology across three Sprints, while functional performance was evaluated using Black Box testing. The test results showed a 100% success rate across 12 functional test scenarios, including item management, QR Code scanning for transaction logging, and automated report generation in PDF and Excel formats. The implementation of the Scrum framework successfully streamlined the development cycle, providing an integrated platform that enhances stock monitoring accuracy and operational efficiency.

REDUKSI DATA BERLABEL PADA DETEKSI TUBERKULOSIS BERBASIS CITRA X-RAY MENGGUNAKAN FRAMEWORK SIMCLR

Majeri Majeri, Fahrim Irhamna Rachman, Muhammad Faisal — Wed, 09 Sep 2026 03:47:06 +0000

The development of deep learning-based Computer-Aided Diagnosis (CAD) for tuberculosis (TB) detection faces a fundamental challenge: high reliance on massive annotated data, which requires scarce radiological expertise, considerable time, and high costs. This study proposes a Self-Supervised Learning (SSL) approach via the SimCLR framework as a strategy to reduce labeled data requirements in X-Ray-based TB classification. The model utilizes a ResNet-50 encoder trained contrastively on unlabeled data using the NT-Xent Loss, followed by downstream adaptation via linear probing (SimCLR-LP) and fine-tuning (SimCLR-FT). Utilizing datasets from UPF BBKPM Makassar, evaluations were conducted across four labeled data fractions (10%, 25%, 50%, 100%). Results demonstrated that at the 10% fraction, SimCLR-LP achieved 85.50% accuracy and an AUC of 0.9091, significantly outperforming the Baseline model (62.60% accuracy) which suffered from degenerate prediction. The SimCLR-LP variant achieved ≥80% accuracy using only 60 labeled images, whereas the Baseline required 303 images to reach a comparable threshold, demonstrating a fivefold labeling efficiency. Grad-CAM analysis confirmed that SimCLR-FT yielded localized activations in the perihilar and lower lung lobes, unlike the Baseline's scattered activations without anatomical focus.

PENERAPAN RESNET50 DAN SWIN TRANSFORMER PADA IDENTIFIKASI CITRA PENYAKIT DAUN KELAPA SAWIT

Wed, 09 Sep 2026 04:02:08 +0000

This study aims to implement and compare the performance of ResNet50 and Swin Transformer models in classifying palm oil leaf diseases. The decline in palm oil productivity is often caused by disease infections such as Curvularia (leaf spot) and Leaf Rust, necessitating a fast and precise automated identification system. This experimental computational research used a primary dataset of 600 digital images proportionally divided into training, validation, and testing sets. The preprocessing stage included resolution adjustment (resizing), data augmentation to prevent overfitting, and normalization. Model performance evaluation was conducted quantitatively through Confusion Matrix calculations and validated qualitatively through heatmap visualization using the Gradient-weighted Class Activation Mapping (Grad-CAM) method. The test results proved that the ResNet50 architecture outperformed the Swin Transformer with an accuracy of 98.00%, precision of 98.01%, recall of 98.00%, and F1-score of 98.00%, compared to the Swin Transformer's accuracy of 96.00%. Grad-CAM analysis also confirmed that ResNet50 is sharper in specifically localizing local infection areas. Overall, it is concluded that the ResNet50 model is more optimal, stable, and recommended for the palm oil leaf disease classification system in this dataset domain.

KLASIFIKASI PENYAKIT PNEUMONIA MENGGUNAKAN MODEL HYBRID CNN-TRANSFORMER BERBASIS CITRA X-RAY PARU-PARU

Wed, 09 Sep 2026 04:24:57 +0000

This study aims to apply a Hybrid CNN-Transformer model based on Medical Vision Transformer (MedViT) for pneumonia classification using chest X-Ray images. The dataset consisted of 450 images, including 150 pneumonia images, 150 non-pneumonia images, and 150 random images as a control class to test system robustness. The data were obtained from Labuang Baji Hospital, Makassar, during the 2023 to 2025 period. The research stages included data collection, preprocessing, augmentation, dataset splitting, model implementation, training, and performance evaluation. The tested models consisted of CNN, Vision Transformer (ViT), and Hybrid CNN-Transformer. The evaluation used accuracy, precision, recall, F1-score, AUC, confusion matrix, ROC curve, and Grad-CAM visualization. The results showed that the Hybrid CNN-Transformer model achieved the best performance with an accuracy of 95.59%, precision of 96.12%, recall of 95.59%, F1-score of 95.58%, and AUC of 0.9968. The model improved accuracy by 8.83% compared with CNN and produced fewer classification errors. The Grad-CAM visualization also indicated that the model focused on relevant lung areas. These findings indicate that combining CNN local feature extraction with Transformer global context can improve pneumonia classification based on medical images..

PREDIKSI KEBUTUHAN STOK OBAT MENGGUNAKAN METODE HYBRID LONG SHORT-TERM MEMORY (LSTM) DAN CATBOOST

Wed, 09 Sep 2026 04:37:02 +0000

Drug inventory planning in primary health facilities requires an accurate forecasting model because fluctuating demand can trigger stockouts or excess inventory. This study develops and evaluates a hybrid Long Short-Term Memory (LSTM) and CatBoost model for predicting the stock requirements of five essential medicines at Puskesmas Pattingalloang. The dataset consists of monthly drug dispensing records from January 2019 to December 2025. LSTM is applied as a temporal feature extractor with a three-month sliding window, while CatBoost functions as the final nonlinear regression estimator. Model performance is assessed using MAE, RMSE, MAPE, and SMAPE, with a single LSTM model used as the baseline comparison. The results show that model suitability depends on the demand pattern of each medicine. The hybrid LSTM-CatBoost model performs better on highly fluctuating medicines, particularly Paracetamol 500 mg with 24.03% SMAPE and Guaifenesin with 37.28% SMAPE. In contrast, the single LSTM model is more efficient for relatively stable demand, especially Blood Supplement Tablets with 14.09% SMAPE. Forecasting for 2026 also provides annual demand estimates that can support data-driven drug requirement planning. These findings indicate that machine learning-based forecasting is useful for pharmaceutical inventory decision support, but model selection must consider the fluctuation characteristics of each drug.

OPTIMASI PENJADWALAN KULIAH FAKULTAS TEKNIK MENGGUNAKAN PENDEKATAN HIBRIDA GREEDY DAN CONSTRAINT SATISFACTION PROBLEM

Rezki Asriani, Rizki Yusliana Bakti, Titin Wahyuni — Wed, 09 Sep 2026 11:22:38 +0000

Course scheduling is a complex academic process that must consider the availability of lecturers, classrooms, time slots, student groups, and specific regulations of each study program. This study aims to implement the Greedy Algorithm combined with a Constraint Satisfaction Problem (CSP) approach to optimize course scheduling in the Faculty of Engineering. The system was developed using Python and tested using scheduling data from the Even Semester of the 2025/2026 Academic Year at the Faculty of Engineering, Universitas Muhammadiyah Makassar. The dataset consisted of 432 courses from six study programs, 136 lecturers, 16 classrooms, seven operational days, and five class sessions per day. The Greedy Algorithm was employed to generate an initial schedule using a first-fit strategy based on priority and slot availability. Subsequently, CSP was applied to model and resolve constraints, including lecturer conflicts, room conflicts, student conflicts, Zoom-class regulations, MKDU scheduling rules, non-regular class schedules, and fixed PWK schedules. The experimental results indicate that the proposed system successfully generated a conflict-free schedule with a Constraint Satisfaction Rate of 100%. All seven hard constraints were satisfied without any lecturer, room, or student conflicts. The core scheduling process required only 0.679 seconds and successfully resolved nine room conflicts within two iterations. These findings demonstrate that the integration of the Greedy Algorithm and CSP is effective in producing valid, efficient, and adaptive course schedules that meet the academic requirements of the Faculty of Engineering.

ANALISIS OPINI PUBLIK TWITTER TERHADAP ISU ENERGI NASIONAL MENGGUNAKAN PENDEKATAN INDOBERT DAN NAÏVE BAYES

Jennifer Verty, Femmy Johan, Muhammad Rizky Pribadi — Wed, 09 Sep 2026 11:34:22 +0000

The increase in fuel oil prices and electric vehicles has become an issue that continues to develop on Twitter. This study aims to analyze public sentiment regarding fuel oil price increases and Public Electric Vehicle Charging Stations (SPKLU) in Indonesia, comparing the Naive Bayes and IndoBERT methods in sentiment classification. The research method includes tweet data collection, data preprocessing, public sentiment labeling, model training, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The preprocessing stage consists of case folding, tokenization, stopword removal, and stemming of Indonesian-language text. The data consist of 700 tweets collected from Twitter using the keywords related to fuel oil, electric vehicles, and Public Electric Vehicle Charging Stations (SPKLU). The results show that the IndoBERT method has better performance than Naive Bayes in sentiment classification because it is able to understand the context of the Indonesian language. This study is expected to contribute to the development of Indonesian-language sentiment analysis.

SENTIMEN PUBLIK TERHADAP KONTROVERSI BYON COMBAT SHOWBIZ BERBASIS SUPPORT VECTOR MACHINE DAN NAIVE BAYES

Moch. Akbar Ramdani, Mohammad Nur Fitriyadi, Mamok Mamok Andri Senubekti — Wed, 09 Sep 2026 11:48:13 +0000

The controversial match result of the sportainment event Byon Combat Showbiz Vol. 6 sparked massive debates on social media. The high volume of comments filled with slang and combat sports jargon makes public opinion mapping prone to subjective bias and Out-Of-Vocabulary (OOV) problems. This study aims to objectively classify netizens' sentiment polarity and compare the performance of Naïve Bayes and Support Vector Machine (SVM) algorithms. The extracted dataset of 1,081 documents was processed using Custom Dictionary-based Normalization to reduce linguistic noise. Features were extracted using TF-IDF weighting, while class imbalance was handled using the Synthetic Minority Over-sampling Technique (SMOTE) strictly on the training data to prevent data leakage. The results showed that public opinion was dominated by negative sentiments at 64.5%, rooted in criticism of the referee's technical regulations. Model evaluation proved that linear kernel-based SVM had the most optimal performance with an accuracy rate of 59.45%, outperforming Naïve Bayes which reached 58.99%. The use of a custom dictionary and strict test data separation in SVM proved effective in precisely mapping public sentiment on a complex syntactic dataset.

ANALISIS SENTIMEN DAN IDENTIFIKASI PRODUK TERLARIS PADA E-COMMERCE PAKAIAN ANAK MENGGUNAKAN K-NEAREST NEIGHBOR

Siti Jatsiah, Ridha Adjie Eryadi, Mohammad Nur Fitriyadi — Wed, 09 Sep 2026 11:58:06 +0000

The rapid growth of the children's clothing E-Commerce industry requires businesses to understand consumer preferences through digital review data. This study aims to perform Sentiment Analysis of customer reviews and identify best-selling products at the Baju Anak Kece Online Store by utilizing Text Mining techniques. The primary challenge addressed is the volume of unstructured customer reviews, making it difficult for store owners to accurately determine satisfaction levels and purchasing patterns. The method employed in this research is the K-Nearest Neighbor (KNN) algorithm to classify customer reviews into three sentiment categories: Positive, Neutral, and Negative. Text preprocessing stages include case folding, cleaning, tokenizing, Stopword removal, and Stemming using the Sastrawi Library, followed by word weighting using the TF-IDF method. To illustrate the scale of the experiment, this study utilized a dataset of 326 customer reviews, which were validated and divided into an 80% training set and a 20% testing set. The results indicate that the KNN algorithm is capable of classifying customer sentiment with an Accuracy rate of 69.23%. These findings are integrated to identify best-selling products based on the dominance of positive sentiment volume, providing actionable marketing strategy recommendations for the store to optimize stock and enhance competitiveness in the digital market.

IMPLEMENTASI ELLIPTIC CURVE DIGITAL SIGNATURE ALGORITHM (ECDSA) UNTUK VERIFIKASI KEABSAHAN DOKUMEN BERBASIS QR CODE

Paris Al Davie, Ridha Adjie Eryadi, Robby Rohman Sukarya — Wed, 09 Sep 2026 12:52:30 +0000

The ease of modifying digital documents increases the risk of legal data forgery due to the lack of a dual authentication mechanism. This study aims to implement a document validity verification system using the Elliptic Curve Digital Signature Algorithm (ECDSA) integrated into a Quick Response (QR) Code. The system is built on a client-server architecture, where cryptographic computation and public key storage are managed centrally on an application programming interface (API), while the Android application serves as an optical scanner. The validation process applies a dual verification mechanism, namely mathematical verification on the server side and visual verification of the original PDF document. Test results show that the system is capable of completely rejecting data manipulation attempts. The computation time for digital signing remains constant under 1 ms, while the hashing time ranges from 2 ms to 54 ms for document sizes between 500 KB and 5 MB. This system effectively guarantees the integrity and validity of digital documents.

SEGMENTASI NASABAH KREDIT BANK SULSELBAR BERDASARKAN INSTANSI DAN STATUS KEPEGAWAIAN

Aidil Syamsul, Gilang Ghiakhuzuki Ramsidar, Dikwan Moeis, Asri Yadi — Wed, 09 Sep 2026 13:27:19 +0000

Proper management of customer registration data forms the primary foundation for a banking credit division in mapping market potential. This study aims to identify the characteristics and distribution patterns of credit customers at Bank Sulselbar Gowa Branch based on their originating agency and employment status. Using a quantitative descriptive approach, a total of 271 secondary registration data entries were processed through data cleaning, preprocessing, and cross-tabulation stages. The results show that customers with active status dominate the proportion of credit applicants (53.14%), followed by PPPK (21.03%). Based on originating agencies, the Local Government (28.78%) and Elementary Schools (27.68%) are the largest contributors. Cross-tabulation reveals that applicants from the education sector dominate the certification category. This study demonstrates that the efficiency of medium-scale credit administrative data management can be achieved through standardized data processing without relying on complex systems.