cover
Article Per Year (5 Year)
All
Home Page
OAI Link
Editorial Team
Contact
Reviewer
Google Scholar
Contact Name
Budi Hermawan
Contact Email
-
Phone
+62081703408296
Journal Mail Official
info@kdi.or.id
Editorial Address
Jl. Flamboyan 2 Blok B3 No. 26 Griya Sangiang Mas - Tangerang 15132
Location
Kab. tangerang,
Banten
INDONESIA
bit-Tech
Published by Komunitas Dosen Indonesia
ISSN : 2622271X     EISSN : 26222728     DOI : https://doi.org/10.32877/bt
Core Subject : Science,
Computer Science & IT
The bit-Tech journal was developed with the aim of accommodating the scientific work of Lecturers and Students, both the results of scientific papers and research in the form of literature study results. It is hoped that this journal will increase the knowledge and exchange of scientific information, especially scientific papers and research that will be useful as a reference for the progress of the State together.
Arjuna Subject : Ilmu Komputer (semua kategori) - Ilmu Komputer (Lain-Lain)
Articles 642 Documents
 52   53   54   55   56 
Performance Assessment of Machine Learning Models for Gas Turbine Fault Detection Umi Yuliatin; Hernawan Novianto
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3443

Abstract

This study systematically evaluates the effectiveness of classical machine learning models for gas turbine fault detection under class-imbalanced operating conditions. Using an industrial dataset of 1,386 observations with a binary target (30.7% Fault) and nine operational parameters, we first conduct exploratory analysis to characterize correlation structure and extreme operating states. The methodological pipeline comprises stratified train–test splitting, feature standardization, and training-set rebalancing using SMOTE, followed by estimation of four models: Logistic Regression, Random Forest, XGBoost, and Support Vector Machine. Model performance is assessed using common classification indicators, focusing on the trade-off between overall discrimination and the ability to correctly identify Fault conditions. The results show consistently weak discriminative power, with AUC values only slightly above random classification (0.48–0.55) and low effectiveness in detecting Fault cases, despite reasonable accuracy for No Fault conditions. These findings provide an empirical baseline showing that, for this dataset, classical models struggle to achieve clinically meaningful separation between normal and faulty turbine states. The study’s main contribution is to demonstrate, on real industrial data, how limited feature informativeness, class imbalance, and potential label or measurement noise jointly constrain learnability, even after standard rebalancing. A key implication is that reliable gas turbine fault detection will require richer, domain-informed feature engineering particularly temporal and condition-specific descriptors and possibly more expressive models, such as deep learning or hybrid physics-informed approaches. Future research should validate these insights on larger multi-plant datasets and systematically compare advanced feature-learning strategies and cost-sensitive optimization schemes empirically.
Implementation of the KNN Algorithm for Food Recommendation System Based on Users' Nutritional Needs Silvia Mairiani Rosdilillah; Agus Suhendar
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3449

Abstract

This study develops a web-based food recommendation system using the K-Nearest Neighbors (KNN) algorithm to provide personalized food recommendations based on users' nutritional needs and preferences. Many individuals struggle to create balanced diets due to insufficient knowledge or time, which can lead to malnutrition or obesity. To address this, the system calculates users' nutritional needs using Basal Metabolic Rate (BMR) and Total Daily Energy Expenditure (TDEE), incorporating preference filtering provided by users. The KNN algorithm then analyzes a food database to identify items that best match the users' nutritional profiles. The system features two primary interfaces: a user interface for inputting nutritional data and displaying recommendations, and an administrative interface for managing food data, user information, and recommendation history. The system was evaluated through Black Box Testing, which confirmed that all main features function as intended. The KNN algorithm demonstrated effectiveness by providing relevant food recommendations that align with users' individual nutritional requirements. Key evaluation metrics, such as recommendation accuracy and user satisfaction, validate the system's performance. This approach highlights the system’s potential in offering personalized nutrition advice, with a focus on real-time decision-making. Future work will aim to incorporate additional dietary factors and expand the food database to enhance the system’s adaptability and precision.
Thyroid Disease Classification Using Support Vector Machine and Recursive Feature Elimination Method Citra Wulandari; lis Afrianty; Elvia Budianita; Siska Kurnia Gusti
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3454

Abstract

Thyroid disease is a common endocrine disorder that can cause serious metabolic and cardiovascular complications, so accurate early detection is clinically essential. This study proposes a Support Vector Machine (SVM) classifier enhanced with Recursive Feature Elimination (RFE) to select the most informative attributes and Adaptive Synthetic Sampling (ADASYN) to handle class imbalance in a Kaggle thyroid dataset of 3,771 clinical records. The data contain 25 diagnostic attributes with a strongly skewed distribution between healthy and thyroid cases. The model’s robustness was examined using three train–test split ratios. The best configuration, SVM with a Linear kernel and 20 RFE-selected features under an 80:20 split, achieved 98.39% accuracy, with precision, recall, and F1-score all reaching 0.98, indicating consistently strong performance across classes. RFE contributes by removing redundant or weakly relevant variables, helping the classifier construct a more stable and interpretable decision boundary. ADASYN further improves the representation of the minority class, yielding higher recall and F1-score for thyroid cases and reducing the risk of missed diagnoses. Overall, the combined use of feature selection and adaptive oversampling produces a balanced and computationally efficient model for thyroid disease classification. These findings suggest that the proposed approach can support clinical decision-making, reduce diagnostic errors in imbalanced data settings, and strengthen early detection efforts in endocrine health assessment. By offering high sensitivity for thyroid cases while maintaining robust specificity for healthy patients, the model is well suited for integration into clinical decision-support and routine screening workflows.
Application of Backpropagation Neural Network Using Random Oversampling and Robust Scaler for Classification Thyroid Ummy Agustina Putri; Iis Afrianty; Elvia Budianita; Fadhilah Syafria
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Thyroid disease is a fairly common endocrine disorder that requires rapid and accurate diagnosis so that patients can receive appropriate treatment. This study was conducted to improve the system's ability to classify thyroid disease by utilizing data preprocessing techniques with RobustScaler and Random Over Sampling (ROS), as well as the Backpropagation Neural Network (BPNN) algorithm. The research dataset consisted of 3,771 patient data with 25 clinical attributes describing the condition and function of the thyroid. The data preprocessing process involved data selection, data cleaning, and data transformation using RobustScaler so that each feature had a more stable scale and was not affected by extreme values. The class imbalance problem was overcome using ROS so that the amount of data increased to 6,834 samples and the class distribution became more balanced. The Backpropagation Neural Network algorithm was applied in model training by testing various variations in the number of neurons in the hidden layer (38 and 49) and learning rate (0.01 and 0.001). Training was conducted for 5,000 and 10,000 epochs. Evaluation was performed using the 10-Fold Cross Validation method to obtain more consistent results. The results of the study show that the model is capable of achieving very high accuracy, up to 99.85%, on several parameters. The results show that proper data processing and appropriate parameter selection greatly affect model performance. Overall, the use of RobustScaler and ROS has been proven to significantly improve the accuracy of thyroid disease classification.
Comparative Analysis of Dijkstra and A* Algorithms for Determining the Shortest Route Ardiansyah Ardiansyah; Abdul Muin Nasution; Muhammad Iqbal
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3474

Abstract

This study presents a comparative analysis of Dijkstra and A* algorithms for determining the shortest route in an urban road network scenario, specifically from SMKN 9 Medan to Gramedia Gajah Mada, Medan. The road network is modeled as a weighted graph, where nodes represent key locations, and edges represent inter-node distances derived from Google Maps. Three alternative routes are evaluated based on inter-node distances and direct heuristic distances to the destination. Dijkstra’s algorithm, an uninformed search method, guarantees optimality by exhaustively exploring all possible paths with non-negative weights. In contrast, the A* algorithm incorporates a heuristic function that estimates the remaining distance to the goal, enhancing search efficiency by focusing on the most promising paths. Both algorithms are applied to the same graph data for a fair comparison, with performance metrics including total route distance, number of nodes explored, and computational efficiency. The results show that while both algorithms identify the same optimal route (A–B–E–G, 5.7 km), A* outperforms Dijkstra in terms of computational efficiency, exploring fewer nodes and requiring less computation time. These findings suggest that while Dijkstra remains reliable for smaller networks, A* is better suited for real-world navigation applications where efficiency and scalability are critical. This study provides empirical evidence supporting the use of heuristic-based algorithms in urban route planning systems.
Classification of Phishing URL Attacks Using Random Forest Algorithm Based on Feature Importance Melyana Hasibuan; Rahmad Abdillah; Surya Agustian; Reski Mai Candra
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3511

Abstract

The development of information technology and increasing digital activities have made URL-based phishing threats more complex and difficult to detect. Phishing attacks target not only individuals but also organizations, requiring detection systems that are accurate, efficient, and capable of handling high-dimensional data. Machine learning approaches, particularly Random Forest, have been widely applied for phishing detection; however, further evaluation is needed regarding the role of feature selection in improving efficiency without reducing performance. This study aims to evaluate the performance of the Random Forest algorithm for phishing URL detection and to analyze the impact of feature selection based on feature importance. This research adopts the Knowledge Discovery in Databases (KDD) framework, including data selection, preprocessing, feature selection, modeling, and evaluation stages. The PhiUSIIL-2024 dataset is used, with two modeling scenarios: Random Forest using all features (RF Full) and Random Forest using the top 30 features selected through feature importance (RF Top-30). Model performance is evaluated using accuracy, precision, recall, and F1-score metrics under different data split ratios. The experimental results show that both models achieve very high and stable classification performance, with evaluation metrics close to or reaching 100%. The RF Top-30 model maintains performance comparable to the RF Full model despite using fewer features. This study concludes that feature importance-based feature selection effectively simplifies the Random Forest model without sacrificing performance, making it suitable for efficient URL phishing detection systems.
Classification of Sentiment of Emina Product Reviews Using the Naive Bayes Algorithm Wiwik Astriani; Otong Saeful Bachri; Bambang Irawan
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3554

Abstract

The rapid development of e-commerce in Indonesia has led to an increase in the number of consumer reviews containing opinions and experiences of using products. In the cosmetic product category, text reviews have an important role in influencing purchasing decisions. However, the large volume of data and the imbalance of sentiment distribution are the main challenges in conducting manual and accurate sentiment analysis. Therefore, an automated approach based on machine learning is needed that is efficient and capable of handling large-scale and unbalanced data. This study aims to analyze the sentiment of reviews of Emina brand cosmetic products on the Tokopedia platform and evaluate the effectiveness of the Multinomial Naïve Bayes algorithm combined with TF-IDF and SMOTE data balancing techniques in classifying positive, neutral, and negative sentiments. The research data was obtained through web scraping of Emina product reviews, resulting in 446,325 review data. The research stages include text preprocessing, rule-based sentiment labeling, feature extraction using TF-IDF, data balancing using SMOTE, and classification modeling with the Naïve Bayes Multinomial algorithm. Model performance evaluation was carried out using accuracy, precision, recall, F1-score, and confusion matrix metrics. The test results showed that the model achieved an accuracy of 94.72% with a stable F1-score value in all sentiment classes, including minority classes, after the implementation of SMOTE. This study proves that the combination of Multinomial Naïve Bayes, TF-IDF, and SMOTE is effective for large-scale analysis of cosmetic product review sentiment and is able to significantly overcome the problem of data imbalance.
Implementation of Computer Vision for Traffic Light Systems Using Convolutional Neural Networks with YOLOv3 Loe Ju; Joko Susilo
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3555

Abstract

Rapid urban motorization has intensified congestion at signalized intersections, where conventional fixed-time control fails to accommodate fluctuating traffic demand. This study proposes an interpretable, real-time adaptive traffic signal system that integrates deep learning–based perception with fuzzy logic decision-making. Unlike prior works that treat detection and control as separate components, this research establishes an end-to-end perception-to-decision pipeline linking YOLOv3-based vehicle detection to a Mamdani fuzzy inference controller. Traffic videos are processed frame by frame to detect and count vehicles, from which lane-level parameters—vehicle count, queue length, and density—are extracted as fuzzy inputs. The controller adaptively determines green-phase durations according to real-time traffic states. Experiments using 300 real-world video frames under varying congestion conditions achieved precision and recall rates of 0.91 and 0.88, respectively, confirming YOLOv3’s suitability for urban traffic environments. The adaptive system produced dynamic green times ranging from 20 to 52 seconds, reducing average green duration by approximately 29% relative to fixed-time control while maintaining effective queue clearance. These findings demonstrate that the proposed integration achieves both computational efficiency and interpretability, offering a practical alternative to opaque deep reinforcement learning–based controllers. The study contributes to the growing discourse on explainable AI in transportation by operationalizing a transparent, deployable framework that links vision-based sensing to adaptive signal control, enhancing responsiveness and scalability for next-generation intelligent traffic management systems.
Implementation of Business Intelligence to Analyze Product Popularity Manalu, Evant Welsh; Hariyanto, Susanto
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3104

Abstract

PT Dairyfood Internusa is a corporation engaged in the distribution of bakery products and food ingredients to Hotels, Restaurants, Cafes, and retail stores throughout Indonesia. This company conducts numerous transactions annually, resulting in a substantial amount of raw data that can be processed and analyzed to extract important information, which is then presented in visual form. Product popularity is one of many factors that determine the degree to which a product is liked and sought after by customers. It can be influenced by various factors, including its quality, social influence or marketing, and the availability of information about the product. And for this reason, a tool are needed to measure and visualize product popularity which can provide the stakeholder with necessary data that helps them in decision-making. The problem is how to make a system that can extract, transform, and visualize the data statistically in real time. The effort made is to cut labor and time for the users from transforming the raw data, by implementing the 9 Step Kimball methodology for developing a data warehouse which will be used to store the raw data and also the transformed data. Using the application Microsoft Power BI to enabled us to visualize the transformed, the author wants to create and design a business intelligence system that can make it easier for users or stakeholders to see and get the analytical data they needed in decision-making.
An An Explainable Machine Learning Approach Using Random Forest and SHAP for Employee Attrition Prediction Ipmawati, Joang; Kusnawi, Kusnawi
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3281

Abstract

Understanding and predicting employee attrition is a strategic challenge for modern organizations because high turnover rates impact operational costs, productivity, and the loss of valuable company knowledge. Conventional statistical approaches, such as logistic regression, have limitations in capturing complex and non-linear relationships between workforce variables. This study proposes an Explainable Machine Learning approach by integrating the Random Forest algorithm and the SHAP (SHapley Additive Explanations) method to predict and interpret employee attrition behavior more transparently.  However, existing HR analytics research rarely combines tree-based ensemble models with robust explainability, creating a gap in developing accurate yet interpretable solutions.The dataset used is HR-employee-attrition, with 1,470 entries and 35 features covering demographics, compensation, and job satisfaction. After preprocessing and parameter optimization, the Random Forest model achieved 83% accuracy, an ROC-AUC of 0.789, and a PR-AUC of 0.414. Model performance was validated through a 70:30 stratified split supported by cross-validation to ensure predictive consistency, indicating good classification performance despite class imbalance. SHAP analysis identified five key features influencing attrition: OverTime, MonthlyIncome, Age, YearsAtCompany, and JobSatisfaction. Unlike conventional black-box models, the proposed approach provides global and local explanations that clarify the contribution of each feature to individual predictions. Practically, these insights enable HR departments to identify high-risk employees earlier and design targeted retention interventions based on data-driven evidence.The findings demonstrate that integrating Random Forest with SHAP produces models that are both accurate and interpretable. Future research may explore integrating SHAP explanations into interactive HR decision-support systems and evaluating more advanced explainable deep learning methods.

Page 54 of 65 | Total Record : 642

 52   53   54   55   56 

Filter by Year

2018 2026


Reset
Filter By Issues
All Issue Vol. 8 No. 3 (2026): bit-Tech Vol. 8 No. 2 (2025): bit-Tech Vol. 8 No. 1 (2025): bit-Tech Vol. 7 No. 3 (2025): bit-Tech Vol. 7 No. 2 (2024): bit-Tech Vol. 7 No. 1 (2024): bit-Tech Vol. 6 No. 3 (2024): bit-Tech Vol. 6 No. 2 (2023): bit-Tech Vol. 6 No. 1 (2023): bit-Tech Vol. 5 No. 3 (2023): bit-Tech Vol. 5 No. 2 (2022): bit-Tech Vol. 5 No. 1 (2022): bit-Tech Vol. 4 No. 3 (2022): bit-Tech Vol. 4 No. 2 (2021): bit-Tech Vol. 4 No. 1 (2021): bit-Tech Vol. 3 No. 3 (2021): Remote Delivery Vol. 3 No. 2 (2020): Datamining Vol. 3 No. 1 (2020): Distance Learning Vol. 2 No. 3 (2020): Pandemik ICT Vol. 1 No. 3 (2019): Learning Synchronous and Asynchronous in the Industry 4.0 Vol. 2 No. 2 (2019): Support System Vol. 2 No. 1 (2019): Data Mining and Green Technology Vol. 1 No. 2 (2018): Data and Information Quality Vol. 1 No. 1 (2018): First Impression on Computing More Issue