cover
Article Per Year (5 Year)
All
Home Page
OAI Link
Editorial Team
Contact
Reviewer
Google Scholar
Contact Name
Jumanto
Contact Email
jumanto@mail.unnes.ac.id
Phone
+6281339762820
Journal Mail Official
shmpublisher@gmail.com
Editorial Address
Jl. Karanglo Raya No. 64, Pedurungan, Semarang, 50191, Indonesia
Location
Kota semarang,
Jawa tengah
INDONESIA
Journal of Soft Computing Exploration
Published by shm publisher
ISSN : 27467686     EISSN : 27460991     DOI : https://doi.org/10.52465/joscex
Core Subject : Science, Engineering,
Computer Science & IT Control & Systems Engineering Decision Sciences, Operations Research & Management Electrical & Electronics Engineering
The journal focuses on publishing high-quality, original research and review articles in the field of Soft Computing, Informatics and Computer Science, emphasizing the development, application, and rigorous evaluation of Advanced Computational Methods, Artificial Intelligence (AI), Machine Learning (ML), and Data Science to address complex real-world challenges. The scope of the journal includes, but is not limited to, innovative research in the following areas: 1. Artificial Intelligence and Machine Learning Novel Algorithms and Architectures: Development and comparison of ML/DL models for classification and prediction (including Logistic Regression, Ridge Classifier, SVM, k-NN, and Random Forest). Ensemble Learning: Evaluation and optimization of ensemble methods Balanced Random Forest, SMOTE-RF, SMOTEBoost, and RUSBoost for robust prediction. Data Challenges and Preprocessing: Techniques for mitigating issues like class imbalance (using methods like SMOTE and GAN) and feature extraction/dimension reduction techniques (including Principal Component Analysis (PCA) and Local Binary Pattern (LBP)). 2. Deep Learning and Computer Vision Convolutional Neural Networks (CNNs): Research on CNN architectures (VGG16, ResNet50, DenseNet121, EfficientNet, and MobileNetV2) and the impact of optimization functions (Adam, SGD, NAdam) on model performance. Hybrid and Concatenated Architectures: Proposing and evaluating hybrid models (MobileNetV2 combined with LBP) or concatenated architectures (MobileNetV2 and DenseNet201) to improve classification and feature representation. Image Analysis Tasks: Advanced techniques for image classification (specifically Diabetic Retinopathy), image similarity detection (using Siamese Networks and Test-Time Augmentation), and multi-object segmentation (using FCN with Squeeze-and-Excitation and Attention Mechanisms for palm oil images). 3. Data Science and Advanced Analytics Pattern Detection and Data Mining: Performance evaluation of data mining algorithms, including Biclustering (Cheng & Church and Spectral Biclustering), specifically under challenging structural conditions like collinearity and overlap. Time Series Analysis and Forecasting: Application of advanced decomposition and clustering methods (Ensemble Empirical Mode Decomposition (EEMD) and Time Series Clustering with DTW/ARIMA) for accurate economic or temporal prediction. 4. Applied Informatics (Domain-Specific Applications) Health and Medical Informatics: Classification models for disease diagnosis (including Heart Attack Disease and Diabetic Retinopathy). Agricultural Informatics: Automated detection and classification of plant diseases from leaf/crop images (including Mango Leaf Disease and Chili Plant Disease) and Palm Oil Segmentation. Business and Economic Informatics: Predictive modeling for crucial business metrics (Customer Churn Prediction in Telecommunications) and economic forecasting (Rice Price Forecasting).
Arjuna Subject : Ilmu Komputer (semua kategori) - Kecerdasan Buatan
Articles 38 Documents
 1   2   3   4 
Improving intrusion detection performance using bayesian hyperparameter optimization for supervised network traffic classification Dahlan Dahlan; Dadan Saepul Ramdan
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.87

Abstract

The rapid growth of networked systems has increased the complexity of network traffic and the risk of cyber-attacks, making intrusion detection more challenging. Machine learning approaches have been widely used to address this issue; however, their performance often depends on appropriate hyperparameter settings. This study examined the effect of Bayesian-based hyperparameter optimization on the performance of supervised machine learning models for network traffic classification. A publicly available dataset was used, consisting of various traffic-related features and labeled instances indicating normal or malicious activity. Several machine learning models, including Random Forest, Decision Tree, AdaBoost, Logistic Regression, Gradient Boosting, and Naïve Bayes, were evaluated. Each model was tested using default parameters and then optimized using Bayesian Optimization. The performance was assessed using accuracy, precision, recall, and F1-score. The results showed that ensemble-based models, particularly Gradient Boosting and Random Forest, achieved the best performance after optimization, with accuracy values above 89% and strong F1-scores. However, the findings also revealed a trade-off between precision and recall, where higher precision was often associated with lower detection of certain attack instances. In contrast, simpler models such as Logistic Regression showed lower performance, indicating their limitations in capturing complex patterns. Overall, the study demonstrated that Bayesian-based hyperparameter optimization contributed to improving model performance and provided a more reliable approach for network traffic classification.
Integrating vehicle dimension features for vision-based traffic density prediction using YOLOv5-LSTM architecture Filda Angellia; Nita Merlina; Agus Subekti; Rahmadya Trias Handayanto
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.119

Abstract

Traffic congestion in urban areas requires intelligent technology-based solutions to support modern transportation systems. This study proposes a vision-based traffic congestion prediction framework that integrates YOLOv5 with a sequential deep learning model to improve forecasting accuracy. YOLOv5 is used for real-time vehicle detection, while the width and height of the bounding box are extracted as spatial occupancy features to provide additional information beyond conventional vehicle counting methods. Experiments are conducted using six urban traffic videos consisting of 90,012 frames collected under various traffic conditions. The extracted features are converted into sequential temporal records and subsequently used to train Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. Model performance is evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Experimental results show that both models achieve competitive performance for traffic congestion forecasting. LSTM achieved the best performance with an MSE of 3.77, an RMSE of 1.94, and an MAE of 1.47, demonstrating its superior ability to capture long-term temporal dependencies in large-scale sequential traffic data. In contrast, GRU exhibited lower computational complexity and faster inference time due to its simpler architecture. These findings suggest that integrating vehicle dimensional features with sequential deep learning models provides a more effective approach to artificial intelligence.
Image recognition based on combination of RGB–HSV color and shape features using PCA and K-nearest neighbor Hadiq Hadiq; Solehatin Solehatin; Much Aziz Muslim; Dwika Ananda Agustina Pertiwi
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.111

Abstract

This study aims to accurately separate flower objects from complex backgrounds in artificial intelligence (AI)-based plant management systems. Previous studies have shown limitations in preprocessing techniques and often did not explicitly report classification accuracy. To address these issues, the proposed framework consists of eight stages. The process begins with image size standardization to 150 × 150 pixels, followed by low-pass filtering and image sharpening to enhance object boundaries. Segmentation is then performed sequentially using RGB and HSV color models to achieve more precise object separation. Subsequently, Sobel edge detection and thinning are applied to extract geometric features, such as distances between petal tips and flower perimeter measurements. The extracted features are optimized using Principal Component Analysis (PCA), which reduces the original ten attributes to four principal components, thereby eliminating data redundancy before classification using the Euclidean distance-based K-Nearest Neighbor (KNN) algorithm. Experimental results show that PCA preserves 86.7% of the original data variance, while the proposed system achieves an overall average classification accuracy of 88% at k = 5. Specifically, the recognition accuracies obtained for the four flower categories were 93.33% for Flower A, 86.67% for Flower B, 90.00% for Flower C, and 80.00% for Flower D. The main contribution of this research is the integration of intensive preprocessing techniques, the combination of RGB-HSV color features with geometric shape features, and PCA-based feature optimization, which collectively improve the stability and computational efficiency of KNN classification.
Comparative performance of IndoBERT-based deep learning models with SMOTE for trade tariff sentiment analysis Muhammad Rizki Roihan; Muhammad Itqan Mazdadi; Muliadi; Irwan Budiman; Fatma Indriani
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.122

Abstract

The rapid growth of social media discussions on trade tariff policies has produced large volumes of Indonesian-language opinion data, making sentiment analysis an important tool for understanding public responses. However, studies in this domain remained limited, particularly those addressing class imbalance and comparing multiple hybrid architectures. This study aimed to compare four IndoBERT-based model configurations, with and without the Synthetic Minority Over-sampling Technique (SMOTE), for classifying sentiment in Indonesian trade tariff discussions under class imbalance. The dataset consisted of Indonesian tweets related to trade tariffs collected from X (Twitter) between January and July 2025. The configurations were Logistic Regression as a baseline, IndoBERT-BiLSTM, IndoBERT-CNN, and IndoBERT-BiLSTM-CNN. The tweets were passed through IndoBERT to generate contextual embeddings without fine-tuning. Performance was evaluated using accuracy, precision, recall, and macro F1-score, with macro F1-score as the primary metric. The results showed that IndoBERT-CNN with SMOTE achieved the best macro F1-score of 0.7657 and an accuracy of 0.8654. SMOTE consistently improved recall across all deep learning architectures, with IndoBERT-CNN gaining the most, from 0.7304 to 0.7875. These findings showed that, among the four compared configurations, IndoBERT-CNN with SMOTE provided the most balanced performance for imbalanced Indonesian trade tariff sentiment classification.
Hybrid support vector machine and random forest for environmental issue sentiment analysis on social media x Yulius Palumpun; Moh. Rahmat Irjii Matdoan
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.127

Abstract

This study investigates the effectiveness of a hybrid machine learning model combining Support Vector Machine (SVM) and Random Forest (RF) for sentiment analysis of environmental issues discussed on social media X. A quantitative experimental research design was employed using textual data related to environmental topics collected through an application programming interface (API)-based data extraction process. Prior to model development, the collected data underwent a series of preprocessing procedures, including text normalization, tokenization, stopword elimination, and stemming. The processed text was then converted into numerical feature vectors using the Term Frequency–Inverse Document Frequency (TF-IDF) technique. To assess classification performance, three models were implemented and compared: Support Vector Machine, Random Forest, and a hybrid SVM–RF ensemble model. Model evaluation was conducted through cross-validation using accuracy, precision, recall, and F1-score as performance indicators. The experimental results revealed that the hybrid model achieved the best overall performance, attaining an accuracy of 89.10%, compared with 85.30% for Random Forest and 82.45% for Support Vector Machine. In addition, the hybrid approach generated higher precision, recall, and F1-score values, demonstrating greater robustness and consistency in sentiment classification. These findings suggest that integrating multiple machine learning algorithms can significantly enhance the analysis of complex and unstructured social media data concerning environmental issues.  
Clickbait detection in indonesian news headlines using various prompting strategies in large language models Nandhika Rega Rohadi; Fajar Muslim; Dewi Wisnu Wardani
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.131

Abstract

Clickbait detection in news headlines is a critical task in Natural Language Processing (NLP) related to information quality and the credibility of digital journalism. While traditional machine learning and deep learning approaches have demonstrated impressive performance in clickbait detection, they are limited by a heavy reliance on extensive annotated datasets and significant computational requirements for model training. Unlike previous methods, Large Language Models (LLMs) do not require massive amounts of annotated data. LLMs allow classification tasks to be solved through zero-shot and few-shot prompting without additional retraining. However, the effectiveness of these models can depend significantly on prompting configuration. Despite this, linguistically-enriched prompting strategies have not been widely evaluated for non-English domains such as Indonesian news headlines. This study aims to analyze the performance of various LLM prompting strategies in detecting Indonesian-language clickbait headlines. For this purpose, this study evaluated Llama 4 on the CLICK-ID dataset using multiple combinations of plain and linguistically-enriched prompts (zero-shot and few-shot) alongside advanced inference techniques (self-consistency, weighted self-consistency, and Self-Refine). Performance was measured via Accuracy and Macro F1-scores against a fine-tuned IndoBERT as baseline model. The results show that the prompting approach in the Large Language Models (LLMs) can be used to classify Indonesian clickbait effectively. The use of linguistic prompting and few-shot managed to provide the best performance, which achieved an accuracy of 0.90 and a Macro F1-score of 0.89. 
A model for determining stock purchase decisions based on gated recurrent units and decision tree C4.5 Muhammad Arsyad Rayhan Aziis; Wiranto; Arif Rohmadi
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.133

Abstract

The stock market is highly volatile but offers high profit potential. This makes it difficult for novice investors to make investment decisions. Current studies mostly focus on stock price prediction or trading signal classification, while interpretable hybrid frameworks to support stock purchase decisions are still limited. Many machine learning models offer limited interpretability. Numerical predictions are often generated by forecasting models, which are difficult to translate into investment decisions. Therefore, this study aims to design a hybrid decision support system with an interpretable model by combining a Gated Recurrent Unit (GRU) and a C4.5 Decision Tree classifier in stock purchase decision making. The proposed framework consists of two phases. The first step is to predict the closing price of stocks using the historical daily data with the GRU model. The predicted price is then combined with technical indicators such as Simple Moving Average (SMA), Relative Strength Index (RSI) and Moving Average Convergence Divergence (MACD) to produce trading signals using C4.5 Decision Tree. The dataset used in this study is BBNI.JK stock data from 2015 to 2025 with a walk-forward validation scheme and evaluation using RMSE, MAE, MAPE, Accuracy, Precision, Recall and F1-Score. The experimental results show that the MAPE of the GRU model is 6.16% and the proposed DT-GRU strategy produces the highest trading return of 83.07% with a Sharpe ratio of 2.75. These results indicate that the combination of interpretable forecasting and classification can provide effective and practical trading decision-making support.
How few features are enough? A systematic study of feature reduction in system call N-Grams for android malware detection Rajif Agung Yunmar; Andika Setiawan
Journal of Soft Computing Exploration Vol. 7 No. 2 (2026): June 2026
Publisher : SHM Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52465/joscex.v7i2.120

Abstract

System call N-gram representations have been widely utilized in Android malware detection due to their ability to capture behavioral execution patterns. However, previous studies have not systematically examined how aggressive feature reduction affects different N-gram orders, particularly in extremely low-dimensional feature spaces. This study evaluates the impact of feature reduction on system call N-gram representations ranging from 1-gram to 4-gram using Chi-Square feature selection across subsets varying from Top-3 to Top-100 features. Malware classification was performed using the XGBoost algorithm with leakage-free stratified 10-fold cross-validation. Experimental results show that the 2-gram representation achieved the highest accuracy of 96.98% under full-feature conditions. Surprisingly, under aggressive feature reduction, the 1-gram representation consistently outperformed higher-order N-grams across all reduced feature subsets, indicating greater robustness in low-dimensional feature spaces. Statistical significance analysis using paired t-test and Wilcoxon signed-rank test further confirmed significant performance differences between 1-gram and higher-order N-gram representations. Although higher-order N-grams provide richer contextual and more discriminative behavioral representations, their information becomes increasingly fragmented in sparse feature spaces, reducing effectiveness under extreme dimensionality reduction. Furthermore, aggressive feature selection substantially improved computational efficiency, reducing the execution time of the 4-gram representation from 3000.78 seconds to 8.28 seconds under the Top-100 scenario. Overall, the findings reveal a trade-off between contextual richness, feature sparsity, detection performance, and computational efficiency in system call N-gram-based Android malware detection.

Page 4 of 4 | Total Record : 38

 1   2   3   4 

Filter by Year

2026 2026


Reset
Filter By Issues
All Issue Vol. 7 No. 2 (2026): June 2026 Vol. 7 No. 1 (2026): March 2026