Claim Missing Document
Check
Articles

Found 39 Documents
Search

Penerapan PSO-SVM Untuk Deteksi Serangan Web Dengan Pendekatan Hybrid Anomaly-Signature Based Pratama, Novandi Kevin; Junaidi, Achmad; Nurlaili, Afina Lina
Journal Cerita: Creative Education of Research in Information Technology and Artificial Informatics Vol 11 No 1 (2025): Journal CERITA : Creative Education of Research in Information Technology and Ar
Publisher : UNIVERSITAS RAHARJA

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33050/cerita.v11i1.3497

Abstract

The security of web applications is becoming increasingly crucial with the growing use of web platforms in education and business, especially due to the management of sensitive data. Attacks such as SQL Injection often pose serious threats to data integrity by exploiting weaknesses in input validation. Signature-based approaches are employed to detect known attacks, but they are often ineffective against new threats. On the other hand, anomaly-based approaches using Machine Learning can identify anomalous patterns but are typically slow for real-time detection. This study implements PSO-SVM (Particle Swarm Optimization-Support Vector Machine) to enhance the detection of attacks on web applications by combining anomaly and signature-based approaches. PSO is utilized to optimize SVM parameters, aiming to improve the accuracy of detecting new attacks and reduce the number of undetected threats. Evaluation through testing scenarios demonstrated an accuracy improvement of up to 99.3%, confirming that this hybrid approach is effective in enhancing the security of web applications.
OPTIMASI ALGORITMA K-NEAREST NEIGHBOR DENGAN ALGORITMA GENETIKA PADA DETEKSI PENYAKIT DIABETES MELLITUS Darmawan, Marcellinus Aditya Vitro; Haromainy, M. Muharrom Al; Junaidi, Achmad
JATISI Vol 12 No 2 (2025): JATISI (Jurnal Teknik Informatika dan Sistem Informasi)
Publisher : Universitas Multi Data Palembang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35957/jatisi.v12i2.11353

Abstract

This study discusses the optimization of the K-Nearest Neighbor (KNN) algorithm using Genetic Algorithm (GA) in detecting diabetes mellitus. The research includes stages of collecting datasets on diabetes mellitus symptoms, data preprocessing through normalization and dataset alignment, model implementation, and testing with various scenarios to achieve the highest accuracy. The data used consists of the Pima Indians Diabetes Database as dataset 1 and the Early Stage Diabetes Risk Prediction Dataset as dataset 2. The evaluation is conducted by comparing the accuracy results between KNN without optimization and KNN optimized using Genetic Algorithm. The study's results indicate that optimization is performed by finding the optimal combination of the k-value and the features used in classification. The Genetic Algorithm produces individuals with the best fitness based on the combination of k-values and features that yield the highest accuracy. Testing was conducted on two datasets with two different fold values. The best accuracy was obtained in the 10-fold test, where the accuracy for dataset 1 increased from 74.2% to 79.1% after optimization. Meanwhile, for dataset 2, the accuracy improved from 97.5% to 98.2% after optimization. There was an increase in accuracy for dataset 1, whereas for dataset 2, the improvement was not significant. The conclusion of this study is that optimizing the KNN algorithm using Genetic Algorithm has proven to enhance the accuracy of diabetes mellitus detection, especially in numerical datasets with more complex features.
Penerapan Gated Recurrent Unit dengan Bayesian Optimization dalam Prediksi Harga Saham Sektor FMCG Mas Diyasa, I Gede Susrama; Akmal, Mohammad Faizal; Junaidi, achmad
JIEET (Journal of Information Engineering and Educational Technology) Vol. 9 No. 1 (2025)
Publisher : Universitas Negeri Surabaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26740/jieet.v9n1.p36-41

Abstract

Peningkatan partisipasi investor muda terutama dari Generasi Z dan Milenial menciptakan kebutuhan mendesak untuk menggunakan metode prediksi yang lebih akurat guna meminimalkan risiko investasi. Penelitian ini bertujuan untuk mengembangkan model prediksi harga saham pada sektor Fast-Moving Consumer Goods (FMCG) di Indonesia dengan memanfaatkan algoritma Gated Recurrent Unit (GRU) yang dioptimalkan menggunakan teknik Bayesian Optimization. Metode penelitian ini dimulai dengan pembagian data saham PT Hanjaya Mandala Sampoerna Tbk (HMSP) dari tahun 2019 hingga 2025, yang dibagi menjadi data train (60%), data validation (20%), dan data test (20%). Selanjutnya, dilakukan preprocessing data berupa normalisasi dan sequencing untuk mempersiapkan data. Model GRU yang diterapkan diuji dengan menggunakan metrik evaluasi seperti Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), dan Mean Absolute Percentage Error (MAPE), yang menghasilkan akurasi prediksi yang tinggi dengan RMSE 17.07, MAE 11.50, dan MAPE 1.48%. Penelitian ini menunjukkan bahwa penerapan Bayesian Optimization dapat memberikan efektivitas pemilihan hyperparameter menghasilkan model yang lebih presisi dalam memprediksi harga saham FMCG di Indonesia dan memberikan panduan yang lebih andal bagi investor dalam pengambilan keputusan investasi
KLASIFIKASI PERULANGAN KANKER TIROID MENGGUNAKAN STACK ENSEMBLE DAN SMOTE Rahmanda Putri, Endin; Arman Prasetya, Dwi; Junaidi, Achmad
JATI (Jurnal Mahasiswa Teknik Informatika) Vol. 9 No. 3 (2025): JATI Vol. 9 No. 3
Publisher : Institut Teknologi Nasional Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36040/jati.v9i3.13616

Abstract

Kanker tiroid berdiferensiasi (Differentiated Thyroid Cancer/DTC) memiliki tingkat perulangan sekitar 20%, sehingga identifikasi sejak dini menjadi krusial untuk intervensi dan rencana perawatan terhadap kekambuhan. Penelitian ini menggunakan dataset dari UCI Machine Learning Repository yang berisi 17 atribut klinis pasien dengan proporsi data latih dan uji 80:20. Untuk menangani ketidakseimbangan kelas, diterapkan Synthetic Minority Over-sampling Technique (SMOTE). Model Decision Tree, Support Vector Machine (SVM), dan Logistic Regression digunakan sebagai base learner, sementara meta learner dipilih dari salah satu algoritma tersebut untuk membentuk Stack Ensemble Learning. Decision Tree adalah model paling stabil, dengan akurasi 97% baik sebagai model tunggal, dengan SMOTE, maupun sebagai meta learner dalam Stack Ensemble. SVM memiliki akurasi 83% pada dataset asli, yang meningkat menjadi 94% setelah diterapkan SMOTE. Logistic Regression menunjukkan akurasi 96% di semua skenario. Stack Ensemble dengan meta learning Decision Tree dan Logistic Regression mempertahankan akurasi 97%, sedangkan SVM sebagai meta learner menunjukkan penurunan AUC. Analisis kurva ROC (Receiver Operating Characteristics) menunjukkan bahwa Stack Ensemble dan SMOTE meningkatkan AUC untuk Logistic Regression dan Decision Tree, namun SVM sebagai meta learner dengan dataset SMOTE mengalami penurunan performa dengan nilai terendah 0,94. Hasil ini membuktikan bahwa kombinasi Stack Ensemble dan SMOTE efektif dalam menangani ketidakseimbangan data pada dataset Differentiated Thyroid Cancer Recurrence.
KLASIFIKASI TUTUPAN LAHAN PADA CITRA SENTINEL-2 DI KAWASAN IKN MENGGUNAKAN GOOGLE EARTH ENGINE Al Fathoni, Hanif; Junaidi, Achmad; Prima Aditiawan, Firza
JATI (Jurnal Mahasiswa Teknik Informatika) Vol. 9 No. 3 (2025): JATI Vol. 9 No. 3
Publisher : Institut Teknologi Nasional Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36040/jati.v9i3.13652

Abstract

Pemerintah Indonesia telah meresmikan pemindahan Ibu Kota Negara (IKN) ke Nusantara melalui Undang-Undang Nomor 3 Tahun 2022. Nusantara dirancang sebagai simbol identitas nasional dan pusat pertumbuhan ekonomi dengan konsep keberlanjutan. Pemindahan ini berdampak pada tata ruang, infrastruktur, dan lingkungan, sehingga analisis tutupan lahan menjadi krusial untuk memastikan perencanaan yang efisien. Penelitian ini bertujuan untuk mengklasifikasikan tutupan lahan di kawasan IKN menggunakan citra satelit Sentinel-2 dan teknologi Google Earth Engine (GEE). Algoritma yang digunakan adalah Random Forest (RF) dan Support Vector Machine (SVM), dengan ekstraksi fitur berbasis indeks spektral NDVI, NDBI, dan NDWI. Teknik cloud masking dengan QA Band diterapkan untuk meningkatkan kualitas data sebelum analisis lebih lanjut. Tahapan penelitian meliputi pengumpulan dan pre-processing data citra Sentinel-2, ekstraksi fitur, pembuatan dataset latih dan validasi, serta proses klasifikasi menggunakan algoritma RF dan SVM. Evaluasi dilakukan dengan metrik akurasi, presisi, recall, dan F1-score untuk menentukan model terbaik. Hasil penelitian menunjukkan bahwa model RF dengan 100 pohon (RF_100trees) dan SVM dengan kernel linear (SVM_LINEAR) memiliki akurasi validasi terbaik sebesar 88%. RF unggul dalam kestabilannya dengan jumlah pohon yang besar, sementara SVM lebih sensitif terhadap pemilihan parameter kernel. Kesimpulannya, kedua model ini efektif dalam klasifikasi tutupan lahan kawasan IKN.
Cloud-Based High Availability Architecture Using Least Connection Load Balancer and Integrated Alert System Prinafsika; Junaidi, Achmad; Muharrom Al Haromainy, Muhammad
bit-Tech Vol. 8 No. 1 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i1.2520

Abstract

Ensuring optimal service continuity remains a critical challenge in cloud computing, especially when dealing with high traffic loads and system failure potential that can cause losses. To address this, this research presents the implementation of a high availability (HA) cloud system using the Least Connection load balancing algorithm implemented with Nginx, integrated with early anomaly detection and alert mechanisms. The HA architecture is implemented across two geographically distributed cloud service providers, Alibaba Cloud and Google Cloud, to analyze latency and performance differences under high load conditions. The system's resilience and scalability were evaluated through load testing using K6, simulating workloads ranging from 100 to 1000 Virtual Users (VUs) for single server configurations and 200 to 2000 VUs for HA configurations. The experiment results showed a significant improvement in service availability, reaching 100% uptime with the HA configuration compared to a peak of 98.79% in the single server environment. The Least Connection strategy effectively balanced traffic by monitoring active connections, resulting in a 29.73% increase in processed requests and a 42% reduction in system load at 1000 VUs. Additionally, the alert system successfully sent real-time Telegram notifications for delays or failures, enabling proactive mitigation. These results confirm that combining dynamic load balancing with proactive alerts can significantly improve service reliability, resource efficiency, and resilience to failures in distributed cloud infrastructure providing a viable model for robust and scalable cloud service architectures.
Website Security Testing Using PTES Method and OWASP Top 10 Approach Firnanda, Mochammad Yoga; Henni Endah Wahanani; Achmad Junaidi
bit-Tech Vol. 8 No. 1 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i1.2564

Abstract

Rapid technological advancements have greatly benefited the industrial sector, making technology essential for business operations. However, this reliance also introduces vulnerabilities, particularly in Enterprise Resource Planning (ERP) systems, which are critical for managing business processes and sensitive data. Due to their complexity and integration, ERP systems are prime targets for cyberattacks, emphasizing the need for robust security testing. This research aims to identify, evaluate, and exploit vulnerabilities in the ERP website of PT. XYZ, specifically targeting pages accessible by users with the SPV Marketing role. The Penetration Testing Execution Standard (PTES) methodology was used to guide the process from intelligence gathering to exploitation and reporting. PTES also ensures that testing is conducted legally during the pre-engagement phase. Tools such as Google Dorking, Netcraft, Wappalyzer, and Nmap were employed for intelligence gathering. For threat modeling, ISO 27005 was employed to identify vulnerabilities, while ISO 25010 served as a standard for security quality. A ZAP scan revealed 23 security vulnerabilities, including 18 that fall under the OWASP Top 10, such as Broken Access Control and Injection. Simulated attacks successfully identified Cross-Site Scripting (XSS), Session Hijacking, and Cross-Site Request Forgery (CSRF). Based on the findings, the recommendations focus on enhancing ERP system security according to the OWASP Top 10 guidelines, ensuring clarity for the development team. This study highlights the need for improved ERP security and offers a structured PTES-OWASP framework applicable across sectors. Future research may integrate multiple tools to enhance vulnerability detection.
Optimasi Hiperparameter LSTM Menggunakan PSO untuk Peramalan Bawang Merah dan Bawang Putih Tanjung, Mutiq Anisa; Sari, Anggraini Puspita; Junaidi, Achmad
bit-Tech Vol. 8 No. 1 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i1.2569

Abstract

This research develops a shallot and garlic price prediction model using a Long Short-Term Memory (LSTM) network optimized through the Particle Swarm Optimization (PSO) method. Indonesia experiences an annual increase in demand for these two commodities. This research focuses on optimizing LSTM parameters, such as the number of units in each layer, learning rate, batch size, time step, and number of training epochs using PSO. Various trials were conducted with different PSO parameter settings and data partitioning scenarios to find the best configuration in predicting prices. The results show that the LSTM model optimized with PSO produces an RMSE value of 436,969 for shallots and 173,866 for garlic. In addition to RMSE, the Mean Absolute Percentage Error (MAPE) and R² metrics also show high prediction accuracy. The 90:10 data partitioning scenario showed the best evaluation results, indicating that more data improves the accuracy of the LSTM in learning price patterns. Scatter plots comparing predicted prices with actual prices show a good match, although there is some variation in certain price ranges. This study also highlights the effect of data partitioning on model performance. The LSTM-PSO approach proved effective in improving the accuracy of price predictions and has practical implications for farmers and policy makers in decision making. The model has the potential to be a decision support tool in the agribusiness sector, with the possibility of further development with external factors.
Segmentasi Optic Cup dan Optic Disc Menggunakan U-Net Backbone Resnet50 Bachtiar Riza Pratama; Fetty Tri Anggraeny; Achmad Junaidi
Jurnal Informatika Polinema Vol. 11 No. 4 (2025): Vol. 11 No. 4 (2025)
Publisher : UPT P2M State Polytechnic of Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33795/jip.v11i4.7352

Abstract

Glaukoma merupakan penyakit mata serius yang dapat menyebabkan kebutaan permanen. Salah satu indikator penting dalam diagnosis glaukoma adalah nilai Cup to Disc Ratio (CDR), yang diperoleh dari segmentasi area optic disc (OD) dan optic cup (OC) pada citra fundus retina. Penelitian ini mengembangkan model segmentasi berbasis U-Net dengan backbone ResNet50 untuk mendeteksi area OD dan OC secara otomatis. Data yang digunakan adalah dataset REFUGE sebanyak 1200 citra fundus dan mask ground truth. Sebelum pelatihan, dilakukan tahap pra-pemrosesan yang mencakup ekstraksi ROI optic disc menggunakan metode Normalized Cross-Correlation (NCC) dan peningkatan kontras dengan CLAHE.Model dievaluasi menggunakan metrik Dice Coefficient dan Intersection over Union (IoU) untuk mengukur akurasi segmentasi. Hasil segmentasi menunjukkan bahwa model menghasilkan nilai Dice Coefficient sebesar 0,9175 dan IoU sebesar 0,8976 untuk segmentasi optic disc, serta Dice sebesar 0,8924 dan IoU sebesar 0,8057 untuk segmentasi optic cup. Guna memperhalus bentuk kontur, diterapkan metode ellipse fitting pada hasil segmentasi sebelum perhitungan CDR. Nilai CDR yang diperoleh kemudian digunakan untuk mengklasifikasikan tingkat keparahan glaukoma.
Analisa Komparasi Algoritma Machine Learning dan Deep Learning Dalam Klasifikasi Citra Ras Kucing Royan Fajar Sultoni; Achmad Junaidi; Eva Yulia Puspaningrum
Neptunus: Jurnal Ilmu Komputer Dan Teknologi Informasi Vol. 2 No. 3 (2024): Agustus : Jurnal Ilmu Komputer Dan Teknologi Informasi
Publisher : Asosiasi Riset Teknik Elektro dan Informatika Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.61132/neptunus.v2i3.251

Abstract

Cats (Felis catus) are a type of carnivorous mammal from the Felidae family that was domesticated and has been one of the animals that has mingled with humans since time immemorial. Domestic cats are broadly divided into 2 types, namely village cats and purebred cats. Purebred cats have quite a varied number of types. Therefore, confusion often occurs in determining the type or breed of cat. Meanwhile, in practice, each race does not have the same treatment (especially in the aspect of care). In digital image processing, Machine Learning and Deep Learning are the main aspects in the process of applying technology that can overcome this problem, so research related to this problem was designed. This research was conducted to add insight for further research in a more sophisticated and effective image recognition process. In the experiments carried out in this research, the SVM, KNN, and CNN methods were tested with the Xception and EfficientNet-B1 architectures. Based on the final results obtained from this test, the CNN method with the Xception architecture is the best model. By using fine-tuning and a learning-rate of 1e-5, this method produces a micro average value of 0.974, on a cat breed image dataset of 13 classes and 7800 images. Meanwhile, the method that produces the fastest ETA Training and Testing is obtained by the KNN method, with an ETA Training time of 0.194 seconds, and an ETA Testing time of 1.782 seconds.