Jurnal Informatika: Jurnal Pengembangan IT
Not a member yet
437 research outputs found
Sort by
Pengenalan Pola Aksara Ulu Banyuasin dengan Metode Convolutional Neural Network dan Support Vector Machine
No full textPattern recognition of script characters is a challenge in digital image processing. This study classifies Ulu Banyuasin script using a combination of Convolutional Neural Network (CNN) and Support Vector Machine (SVM). CNN with the VGG16 architecture is utilized for feature extraction, while classification is performed using Multi-Layer Perceptron (MLP) and SVM. The dataset undergoes preprocessing to enhance data quality. Experimental results indicate that the VGG16-SVM combination achieves an accuracy of 99%, outperforming VGG16-MLP, which attains 93%. The performance of VGG16-SVM demonstrates the effectiveness of SVM in improving accuracy after CNN-based feature extraction. However, the risk of overfitting must be considered. Strategies such as data augmentation, hyperparameter tuning, and regularization can be employed to enhance model generalization. This method has proven effective in recognizing Ulu Banyuasin script and can be applied to other character recognition systems
Aplikasi Analisis Sentimen Terhadap Isu Pengelolaan Sampah untuk Solusi Sistem Cerdas Sebagai Upaya Mewujudkan Indonesia Sehat
No full textThe issue of waste management in Indonesia has become a crucial problem that significantly affects both public health and the environment. A lack of public awareness and suboptimal management practices can further worsen environmental conditions and health risks. Therefore, a smarter and data-driven approach is required to understand public opinion and encourage greater community participation. Social media reviews provide valuable insights into waste management issues; however, the overwhelming volume of such data makes analysis challenging.To address this, a sentiment analysis application is needed to summarize public opinions on waste management issues and classify them into positive, negative, and neutral sentiments. The goal is to support more accurate decision-making in formulating effective waste management policies.Sentiment analysis using deep learning methods has demonstrated superior performance compared to lexicon-based and traditional machine learning approaches. BERT, as a deep learning method, has proven to be highly effective in handling textual data. Therefore, this study adopts the BERT method.The research method is divided into two stages. The first stage involves building a sentiment analysis model using BERT, while the second stage focuses on software development using the Software Development Life Cycle (SDLC). The first stage produces the best-performing BERT model, which is then applied in developing a sentiment analysis application for waste management issues.The findings of this study indicate that the BERT method is highly suitable for datasets on waste management issues obtained from Twitter crawling, achieving an accuracy of over 90%. The best-performing model was successfully implemented in a web-based sentiment analysis application
A Systematic Review: Aggregation Methods for Production Processes in Supply Chain Management
No full textIn the modern era, technology has significantly changed the way businesses operate, leading to the need for faster and more efficient processes, new technologies like robotics, machine learning, and artificial intelligence. This has enabled organizations to increase operational efficiency, increase customer experience, and remain competitive in the rapidly changing business environment. One strategy for implementing technology is through ideal generation planning, which involves planning the entire production process and operational planning. This approach helps companies optimize resources, reduce costs, and increase efficiency. In business management, aggregate planning is crucial for integrating various business functions, such as sales, production, and financial management, to achieve a company's full potential. However, implementation can be challenging due to various challenges, such as high production volumes and low volume. This study aims to explore the implementation of aggregate planning in business management, focusing on the impact of technology on efficiency and effectiveness of production planning. Based on the results of the analysis in the journal, it was found that production planning is important at the MSME business unit level. Chase Strategy is the right choice for production planning at the MSME business unit level. Information technology integration has proven critical to improving aggregate planning efficiency, although interdepartmental coordination challenges are a major obstacle. Therefore, it is necessary to implement centralized information technology that is able to unite the needs of each department to achieve overall business process efficiency and effectiveness
Comparison of Machine Learning Algorithm for Enzyme Production Optimization from Industrial Waste
No full textThe manufacture of industrial enzymes from trash provides a sustainable remedy for environmental issues. This work investigates machine learning methods to enhance enzyme production from industrial waste by examining critical factors such as waste type and chemical makeup. Three algorithms—Linear Regression, Decision Tree, and Neural Network—were used to estimate and forecast enzyme production. Evaluation criteria, such as Mean Squared Error (MSE) and Coefficient of Determination (R²), were used to evaluate model performance. The results indicated that the Decision Tree method was the most effective, exhibiting lowest error and enhanced accuracy in selecting ideal production factors such as fermentation temperature and time. This method improves efficiency, lowers operating expenses, and encourages sustainable waste management practices. The results highlight the potential of machine learning to convert trash into useful industrial goods, providing a route to more sustainable biotechnology. Future study may enhance hybrid algorithms, include new waste factors, and facilitate real-time implementation for wider industrial applicability.
Pengembangan Prototipe untuk Prediksi Tingkat Penyeduhan Kopi Menggunakan Data Spektroskopi dan Deep Learning
No full textConsistency in coffee flavor is a crucial factor for coffee enthusiasts, thus requiring a method capable of objectively measuring the coffee brewing level in accordance with the standard brewing chart. This study utilizes the AS7265X spectroscopy sensor to capture the characteristics of coffee based on the resulting light spectrum. The spectral data is then used in a deep learning model using the Convolutional Neural Network (CNN) algorithm to classify the coffee brewing level into five distinct classes. A total of 150 data samples were used in the training and testing process. Initial results show that the model achieved a very high average accuracy of approximately 97%. After hyperparameter tuning using the Random Search method, the model's accuracy further improved, reaching a very high accuracy. However, this performance improvement resulted in a trade-off in computational time, with execution time increasing from 15 seconds to 1 minute and 43 seconds. This research is expected to contribute to ensuring consistent coffee brew quality and to open opportunities for further studies that combine sensor technology and artificial intelligence in the food and beverage sector
Identifikasi Penyakit Pada Daun Kelapa Sawit Dengan Pendekatan CNN AlexNet
No full textOil palm is a plant that plays an important role in Indonesia's agricultural commodities. Cultivating oil palm is suitable for Indonesia due to its tropical climate, which greatly supports the growth of this plant. However, cultivating oil palm is not easy. The emergence of leaf diseases in oil palm can hinder growth, thereby affecting fruit production levels. This research aims to identify diseases on oil palm leaves using one of the methods of Deep Learning, namely the Convolutional Neural Network (CNN) method. This method was chosen because CNN leverages image-based datasets for classification and prediction, making it highly suitable for identifying diseases on oil palm leaves. The research begins with collecting a dataset of images of diseased oil palm leaves. The collected dataset will undergo pre-processing to enhance image quality, enabling more efficient processing by the model. The classification results will subsequently be evaluated to determine the accuracy level of the image processing performed by the model. By implementing Convolutional Neural Network, this research is expected to produce an effective and accurate system for identifying diseases on oil palm leaves, assisting farmers in cultivating oil palm, reducing losses, and ultimately increasing the productivity of oil palm plantations
Design Thinking untuk Perancangan UI Website Seller Toko: Studi Kasus PT Vetencode Pradani Abadi
No full textCompetition in the digital era is a significant challenge for traditional shops in neighborhood communities (RT/RW), particularly those with limited access to and understanding of digital technology needed to compete in online markets. This study aims to design a seller website user interface (UI) using the Design Thinking approach through five stages: Empathize, Define, Ideate, Prototype, and Test. Data collection was conducted through interviews, observations, and literature studies to ensure that the design aligns with the needs of small-scale business owners. The UI prototype was developed in the form of low-fidelity and high-fidelity wireframes and tested through usability testing based on the Heuristic Evaluation method by three internal evaluators. The results showed that 60% of the usability principles revealed no issues (severity 0), while the remaining 40% only indicated cosmetic issues (severity 0.33–1.00). The overall average score of 0.23 reflects a very good level of usability without requiring major improvements. Therefore, applying Design Thinking successfully produced a user-friendly UI tailored to neighborhood-level sellers in navigating digital transformation
Comparison of IndoBERT and Bi-LSTM Models for Indonesian Law Violation Text Classification
No full textLegal violations in Indonesia, particularly those under the Criminal Code (KUHP) and the Information and Electronic Transactions Law (UU ITE), are often difficult for the general public to interpret due to the complexity of legal language and article structures. This research aims to build a multilabel classification model that can automatically identify relevant legal articles from user-provided case descriptions. Two models were developed and compared: Bidirectional Long Short-Term Memory (Bi-LSTM) and IndoBERT. Using a manually labeled dataset, both models were evaluated through accuracy, F1-score, and Hamming Loss metrics, as well as 5-fold cross-validation. The results showed that IndoBERT outperformed Bi-LSTM with an average accuracy of 97% and a Hamming Loss of 0.027. However, t-test analysis revealed no statistically significant difference in F1-scores, indicating that both models have comparable effectiveness in capturing multiple labels. A confusion matrix analysis further identified patterns of misclassification in semantically similar articles. This study demonstrates the potential of NLP and deep learning to support legal awareness and provide the public with easier access to legal information
Klasifikasi Tulang Tengkorak Berdasarkan Jenis Kelamin Menggunakan Correlation-Based Feature Selection (CFS) dengan Backpropagation Neural Network (BPNN)
No full textAbstract – In forensic anthropology, sex identification is the initial step in individual identification, with a probability level of 50%, influencing subsequent examinations such as age and height estimation. The skull is the second-best choice after the pelvis for determining sex, with an accuracy of up to 90%. Morphological and metric methods are less reliable due to the high variability of skulls, while DNA analysis is ineffective on burned or damaged bones. Therefore, this study applies Correlation-Based Feature Selection (CFS) with a Backpropagation Neural Network (BPNN) to improve classification accuracy. The dataset used originates from Dr. William Howells, consisting of 2,524 skull samples with 85 variables. CFS was applied with two thresholds, 0.1 and 0.01, and the division of training data and test data using k-fold cross validation with k=10. The BPNN parameters included learning rates of 0.01 and 0.001, along with three different architectures based on the number of input neurons. The results indicate that CFS improved accuracy from 92.06% to 93.25% under the CFS threshold of 0.01, with a learning rate of 0.001 and a BPNN architecture of [72; 95; 1]. This study confirms that combining CFS and BPNN enhances sex classification accuracy based on skull bones.Abstrak – Pada antropologi forensik, identifikasi jenis kelamin adalah langkah awal dalam mengidentifikasi individu dengan tingkat probabilitas 50%, yang berpengaruh pada pemeriksaan lain seperti perkiraan usia dan tinggi badan. Tulang tengkorak menjadi pilihan terbaik kedua setelah tulang panggul dalam menentukan jenis kelamin dengan akurasi hingga 90%. Metode morfologi dan metrik kurang dapat diandalkan karena variabilitas tengkorak yang tinggi, sementara analisis DNA tidak efektif pada tulang yang terbakar atau rusak. Oleh karena itu, penelitian ini menerapkan Correlation-Based Feature Selection (CFS) dengan Backpropagation Neural Network (BPNN) untuk meningkatkan akurasi klasifikasi. Dataset yang digunakan berasal dari Dr. William Howells, terdiri dari 2.524 sampel tengkorak dengan 85 variabel. Pada CFS digunakan dua ambang batas yaitu 0,1 dan 0,01, serta pembagian data latih dan uji data menggunakan k-fold cross validation dengan k=10. Parameter BPNN yang digunakan meliputi learning rate (0,01 dan 0,001) serta tiga arsitektur berbeda sesuai dengan jumlah neuron input. Hasil menunjukkan bahwa CFS meningkatkan akurasi dari 92,06% menjadi 93,25% pada konfigurasi ambang batas CFS 0,01 dengan learning rate 0,001 dan arsitektur BPNN [72; 95; 1]. Penelitian ini menunjukkan bahwa kombinasi CFS dan BPNN dapat meningkatkan akurasi klasifikasi jenis kelamin berdasarkan tulang tengkorak
Prediksi Harga Rumah di Bandung 2024 Menggunakan Ensemble Learning: Analisis Komparatif dan Interpretabilitas
No full textHouse price prediction plays a crucial role in investment decision-making and financial planning, particularly in developing cities like Bandung with its complex property market dynamics. This study aims to evaluate and compare the performance of various ensemble learning techniques in predicting house prices in Bandung for the year 2024, with a specific focus on model interpretability analysis. The data was collected through web scraping from www.rumah123.com in March 2024, covering attributes such as location, number of rooms, land area, and building area. The evaluated ensemble techniques include Random Forest, Gradient Boosting Machines, Xtreme Gradient Boosting, Linear Regression, and Stacking Ensemble. Model performance was assessed using MAE, RMSE, and R-squared metrics, while interpretability analysis was conducted using SHAP values. The Model Stacking Ensemble shows the most optimal results with R² 0.9076, RMSE 0.311, and MAE 0.216 in experiments involving location features. Features such as land size, building size, and location have proven to have the greatest impact in predicting prices based on SHAP analysis. This model has been successfully integrated into a Flask website for interactive price predictions