cover
Contact Name
Huzain
Contact Email
huzain.azis@umi.ac.id
Phone
+628114484875
Journal Mail Official
ijodas.journal@gmail.com
Editorial Address
Jln. Paccerakkang, Kel. Berua, Kec.Biringkanaya, Kota Makassar, Propinsi Sulawesi Selatan, 90241
Location
Unknown,
Unknown
INDONESIA
Indonesian Journal of Data and Science
Published by yocto brain
ISSN : -     EISSN : 27159930     DOI : -
Core Subject : Science, Education,
IJODAS provides online media to publish scientific articles from research in the field of Data Science, Data Mining, Data Communication, Data Security and Data Representation
Articles 195 Documents
A Comparative Study of LSTM and CNN Models in SQL Injection Attack Detection Abdul Rachman Manga'; Wahyu Kadri Rahmat Suat Suat; Huzain Azis
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : Yocto Brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.460

Abstract

Introduction: SQL Injection (SQLi) remains a critical cybersecurity threat because it exploits vulnerabilities in user input validation and can compromise the confidentiality, integrity, and availability of information systems. This study compares Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) architectures for detecting malicious SQL queries under identical experimental conditions. Method: A publicly available dataset containing 148,327 malicious and benign SQL query instances was preprocessed through missing-value removal, label encoding, tokenization, sequence transformation, padding, and embedding representation. LSTM and CNN models were evaluated using three train-test split scenarios of 70:30, 80:20, and 90:10. Performance was assessed using accuracy, precision, recall, F1-score, and confusion matrices, with the 80:20 split selected for detailed evaluation. Results and Discussion: LSTM consistently achieved higher accuracy across the evaluated splits, ranging from 97.84% to 98.00%. Under the 80:20 configuration, LSTM achieved 97.86% accuracy, 99.34% precision, 96.55% recall, and a 97.92% F1-score, compared with CNN at 97.00%, 97.68%, 96.56%, and 97.12%, respectively. LSTM also reduced false positives from 356 to 99, demonstrating better discrimination between legitimate and malicious queries. Conclusion: LSTM provides more reliable SQL Injection detection than CNN by better capturing sequential dependencies within SQL query structures, making it a promising approach for practical cybersecurity systems
Handling Imbalanced Data in Concept Map Proposition Quality Classification: A Comparative Study of Resampling and Class-Weighted SVM Nurul Rismayanti; Lukman Syafie; A Sinra; Rahmadani Rahmadani; Roesman Ridwan Raja
Indonesian Journal of Data and Science Vol. 7 No. 1 (2026): Indonesian Journal of Data and Science
Publisher : Yocto Brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i1.482

Abstract

Introduction: Concept maps are widely used in education to represent students’ knowledge structures, while automatic proposition-level quality classification can support faster and more consistent assessment. However, imbalanced class distributions may cause classification models to favor majority classes and reduce performance on minority categories. Method: This study compared several imbalance-handling techniques for concept map proposition quality classification using Support Vector Machine (SVM) with TF-IDF features. The dataset comprised 691 propositions grouped into four quality classes, with class 3 dominating the distribution. The evaluated approaches included baseline training, Random OverSampler, SMOTE, ADASYN, SMOTE-Tomek, SMOTE-ENN, and cost-sensitive SVM with class_weight='balanced'. Data were divided using an 80:20 stratified split, with balancing applied only to the training set. Results and Discussion: The cost-sensitive SVM achieved the best overall performance with 0.8417 accuracy, 0.7098 macro precision, 0.7320 macro recall, 0.7149 macro F1-score, 0.8455 weighted F1-score, and 0.2158 MAE. Among resampling approaches, ADASYN and Random OverSampler were the most competitive, whereas SMOTE-based variants produced lower macro-level performance. Conclusion: Class-weighted SVM is more effective than the evaluated resampling techniques for sparse, imbalanced concept map proposition data, highlighting the importance of macro-level and ordinal-aware metrics in educational text classification.
Survey-Aware Self-Training Under Label Scarcity for Low-Hemoglobin Screening in NHANES Irawati Irawati; Lilis Nur Hayati
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : Yocto Brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.517

Abstract

Introduction: Semi-supervised learning may reduce labeled-outcome requirements, but confident pseudo-labels can propagate error. We assessed survey-aware self-training for screening smoking-adjusted low hemoglobin using NHANES August 2021–August 2023. Method: The measured cohort included 5,731 adults, of whom 655 met the outcome definition. Predictors comprised demographic, anthropometric, and blood-pressure measures. Outcomes were deliberately withheld from measured respondents after outcome-stratified splitting and mask construction. Five masks provided each fitted model with 20% of training labels and 20% of validation labels; shared tuning used labels from all five masks. Naturally missing-hemoglobin respondents were reserved for descriptive diagnostics. The primary contrast was survey-weighted test average precision (AP) for self-training minus matched supervised boosting, with a 2,000-replicate design-and-mask bootstrap interval. Results and Discussion: The weighted test prevalence was 0.0990. Mean AP was 0.1555 for supervision and 0.1539 for self-training (difference, -0.0016; 95% CI, -0.0076 to 0.0024). Mean Brier scores were 0.08969 and 0.08986 (difference, 0.00017; 95% CI, -0.00105 to 0.00188). Four masks admitted no pseudo-labels and produced identical predictions; the fifth admitted 258 and reduced AP by 0.0080. Conclusion: The comparison was inconclusive. These findings concern the evaluated boosting and admission rules, and the per-model budgets do not represent the total outcome information used in study design and tuning
Leakage-Aware BISINDO Alphabet Classification with Hand-Region DINOv2 Embeddings Lilis Nur Hayati; Dolly Indra
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : Yocto Brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.528

Abstract

Introduction: Visual similarity across images can bias the evaluation of Indonesian Sign Language (BISINDO) alphabet classifiers. This study assessed frozen DINOv2 embeddings and illumination correction using a group-disjoint evaluation protocol. Method: The dataset comprised 19,748 annotated gesture regions from 19,746 images across 26 classes. Full-frame visual-similarity groups were assigned to training, validation, and test partitions containing 14,107, 2,821, and 2,820 instances, respectively. Frozen DINOv2-small embeddings with class-balanced multinomial logistic regression were evaluated under raw, fixed-gamma, and adaptive-gamma conditions against a grayscale reference using principal component analysis and logistic regression. Configuration selection used validation macro-F1, while uncertainty was estimated using 2,000 group-bootstrap resamples. Results and Discussion: The validation-selected raw configuration achieved test accuracy of 90.28% and macro-F1 of 0.8982 (95% CI, 0.8832–0.9130), compared with macro-F1 of 0.4383 for the grayscale reference. Adaptive gamma achieved macro-F1 of 0.9038, but its advantage over raw inputs was small and uncertain (difference, 0.0056; 95% paired CI, −0.0048 to 0.0152). Exploratory brightness tests indicated condition-specific responses to synthetic illumination changes. Conclusion: Frozen DINOv2 embeddings supported static BISINDO gesture-region classification within an auditable evaluation framework. Adaptive gamma did not demonstrate a consistent improvement, and the findings remain limited to annotation-defined regions under the documented grouping protocol.
Residual-Energy LEACH-PSO Extends Simulated WSN Lifetime under Uneven Node Density Ramdan Satra; Muhammad Hattah Fattah
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : Yocto Brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.529

Abstract

Introduction: Uneven spatial node density can concentrate communication burdens in wireless sensor networks (WSNs). This study evaluated the incremental effect of residual-energy cluster-head reassignment on simulated network lifetime. Method: LEACH (M1), particle swarm optimization (PSO) seed clustering (M2), and residual-energy LEACH-PSO (M3) were evaluated across 270 synthetic topologies covering nine near–far scenarios with 30 placements each, producing 810 method runs. Every topology contained 200 nodes in a fixed 200 m × 200 m field. M2 and M3 shared radio, traffic, and PSO settings. Within each logical round, M3 retained the PSO-derived membership and selected the member with the highest residual energy after status transmission as cluster head. The primary topology-paired M3–M2 comparison assessed first-node-dead (FND) and half-nodes-dead (HND) rounds using block-bootstrap confidence intervals. Results and Discussion: Mean FND and HND were 18.70 and 39.85 rounds for M3, compared with 8.84 and 32.84 for M2. Aggregate paired gains were 9.86 FND rounds (95% CI, 9.44–10.26) and 7.01 HND rounds (95% CI, 6.83–7.19). Both endpoints improved on average in all nine scenarios, with the largest gains when 180 nodes occupied the far zone. No endpoints were censored. Conclusion: Cluster-head reassignment based on residual energy extended simulated WSN lifetime under the specified radio and traffic model. Performance under realistic communication conditions and physical deployments requires further validation.