JISKA (Jurnal Informatika Sunan Kalijaga)

Optimizing K-Means Algorithm Using the Purity Method for Clustering Oil Palm Producing Regions in North Aceh

Novia Hasdyna, Rozzi Kesuma Dinata, Balqis Yafis — Fri, 31 Jan 2025 00:00:00 +0700

The K-Means algorithm is a fundamental tool in machine learning, widely utilized for data clustering tasks. This research aims to improve the performance of the K-Means algorithm by integrating the Purity method, specifically focusing on clustering regions renowned for oil palm production in North Aceh. Oil palm cultivation is a vital agricultural sector in North Aceh, contributing significantly to the local economy and employment. This study examines two clustering techniques: the conventional K-Means algorithm and an optimized version, Purity K-Means. Integrating the Purity method increases K-Means' efficiency by decreasing the required convergence iteration. The data used for clustering analysis is sourced from the Department of Agriculture and Food in North Aceh Regency and pertains to oil palm production in 2023. The findings indicate that the Purity K-Means approach notably reduces the iteration count and improves cluster quality. The average Davies-Bouldin Index (DBI) for standard K-Means is 0.45, whereas the Purity K-Means method lowers it to 0.30. Furthermore, applying the Purity method reduced the number of K-Means iterations from 15 to just 3. These results highlight an enhancement in clustering performance and overall efficiency.

Predicting Olympic Medal Trends for Southeast Asian Countries Using the Facebook Prophet Model

Fri, 31 Jan 2025 00:00:00 +0700

The Olympics is a world sporting event held every four years and is a meeting place for all athletes worldwide. The Olympics are held alternately in different countries. The Olympics were first held in Athens in 1896 and have now reached the 33rd Olympics, which will be held in Paris in 2024. Much work has been done to develop prediction models emphasizing improving accuracy to predict Olympic outcomes. However, low-performance regression algorithms are the main problems with prediction. By integrating custom seasonality with the Facebook-Prophet prediction model, this study aims to increase the accuracy of Olympic prediction. The proposed new model involves several steps, including preparing the data and initializing and fitting the Facebook-Prophet model with several parameters such as seasonal mode, annual seasonality, and prior scale. The model is tested using the Olympic dataset (1994–2024). The evaluation results show that this prediction model can provide a good value in predicting the total medals earned. On the Olympic Games (1994-2024) dataset, the model has a very low error MAE, MSE, and RMSE and has an R2 score of 0.99, which is close to perfect. This research shows that the model is effective in improving prediction accuracy.

Implementation of Long Short-Term Memory for Chili Price Prediction in East Java Province

Fata Nabil Fikri, Nurochman Nurochman — Fri, 31 Jan 2025 00:00:00 +0700

Groceries prices often experience fluctuations in several regions in Indonesia, such as East Java Province and one of the commodities is chilies, both red chilies and rawit chilies. Predictive steps that utilize machine learning such as Long-Short Term Memory (LSTM) can be taken to estimate the next price of chili with expectations that the appropriate strategy can be taken by the authorities. LSTM is a network that developed from RNN networks in previous times by offering a longer cell memory so that more information can be stored. This research focuses on finding out whether the LSTM network can be applied to the case of chili price prediction and what architecture and hyperparameter configuration is appropriate for this case. For this reason, the experimental method is used by testing several predetermined variables to obtain the right architecture and hyperparameter configuration. The results of this research show that the LSTM network can be applied in this case and the architecture and best hyperparameter configuration obtained are the same for both types of chilies, namely red chilies and rawit chilies. For red chili, the best RMSE value that can be produced is 1751.890 and 1888.741 for rawit chili.

Extreme Gradient Boosting Model with SMOTE for Heart Disease Classification

Fri, 31 Jan 2025 00:00:00 +0700

Heart disease is one of the leading causes of death worldwide. According to data from the World Health Organisation (WHO), the number of victims who die from heart disease reaches 17.5 million people every year. However, the method of diagnosing heart disease in patients is still not optimal in determining the right treatment. Along with technology development, various models of machine learning algorithms and data processing techniques have been developed to find models that can produce the best precision in classifying heart disease. This research aims to develop a machine learning algorithm model in classifying heart disease to improve the effectiveness of diagnosis and help in determining the right treatment for patients. This research also aims to overcome the limitations of accuracy in existing diagnosis methods by identifying models capable of providing the best results in processing and analysing health data, especially in terms of heart disease classification. In this study, the XGBoost model was identified as the most superior, with an accuracy of 99%. These results show that the XGBoost model has a higher accuracy rate than previous methods, making it a promising solution to improve the accuracy of future heart disease diagnosis and classification.

Class Weighting Approach For Handling Imbalanced Data On Forest Fire Classification Using EfficientNet-B1

Arvinanto Bahtiar, Muhammad Ihsan Prawira Hutomo, Agung Widiyanto, Siti Khomsah — Fri, 31 Jan 2025 00:00:00 +0700

Wildfires threaten ecosystems and human safety, necessitating effective monitoring techniques. Detecting forest fires based on images of forest conditions could be a breakthrough. But, the model built from imbalanced data leads to low accuracy. This research addresses the challenge of class imbalance in multiclass classification for forest fire detection using the EfficientNet-B1 model. This research explores the implementation of class weighting to enhance model performance, particularly focusing on minority classes, namely: Fire and Smoke. A dataset of 7,331 training images was categorized into four classes. The results showed that employing the class weighting method achieved an accuracy of 90%. The training duration of 14 minutes and 45 seconds outperforms the data augmentation method in terms of time efficiency. This study contributes to the development of more effective methods for forest fire monitoring and provides insights for future research in machine learning applications in environmental contexts.

Application of SMOTE in Sentiment Analysis of MyXL User Reviews on Google Play Store

Badriyah Badriyah, Totok Chamidy, Suhartono Suhartono — Fri, 31 Jan 2025 00:00:00 +0700

Texts that express customer opinions about a product are important input for companies. Companies obtain valuable information from consumer perceptions of marketed products by conducting sentiment analysis. However, real-world text datasets are often unbalanced, causing the prediction results of classification algorithms to be biased towards the majority class and ignore the minority class. This study analyzes the sentiment of MyXL user reviews on the Google Play Store by comparing the performance of the Logistic Regression and Support Vector Machine algorithms in the SMOTE implementation. This analysis uses TF-IDF to extract feature and GridSearchCV to optimize the accuracy, precision, recall, and F1 score evaluation metrics. This study follows several scenarios of dividing training data and test data. SVM implementing SMOTE is the algorithm with the best performance using the division of training data (90%) and test data (10%), resulting in accuracy (73.00%), precision (67.13%), recall (65.82%) and F1 score (66.30%).

Revitalizing Art with Technology: A Deep Learning Approach to Virtual Restoration

Nurrohmah Endah Putranti, Shyang-Jye Chang, Muhammad Raffiudin — Fri, 31 Jan 2025 00:00:00 +0700

This study evaluates CycleGAN's performance in virtual painting restoration, focusing on color restoration and detail reproduction. We compiled datasets categorized by art styles and conditions to achieve accurate restorations without altering original reference materials. Various paintings and those with a yellow filter, to create effective training datasets for CycleGAN. The model utilized cycle consistency loss and advanced data augmentation techniques. We assessed the results using PSNR, SSIM, and Color Inspector metrics, focusing on Claude Monet's Nasturtiums in a Blue Vase and Hermann Corrodi's Prayers at Dawn. The findings demonstrate superior color recovery and preservation of intricate details compared to other methods, confirmed through quantitative and qualitative evaluations. Key contributions include employing CycleGAN for art restoration, model evaluation, and framework development. Practical implications extend to art conservation, digital library enhancement, art education, and broader access to restored works. Future research may explore dataset expansion, complex architectures, interdisciplinary collaboration, automated evaluation tools, and improved technologies for real-time restoration applications. In conclusion, CycleGAN holds promise for digital art conservation, with ongoing efforts aimed at integrating across fields for effective cultural preservation.

Comparison of KNN and Random Forest Algorithms on E-Commerce Service Chatbot

Fri, 31 Jan 2025 00:00:00 +0700

Technology heavily influences our lives, with the expansion of e-commerce being an important outcome that demands attention. Given the prevalence of smartphones equipped with messaging apps and fast networks, people often utilize these platforms to communicate with sellers, offering a convenient way for sellers to engage efficiently with a diverse customer base. Recognizing this trend, there is a need for digital transformation of services to improve operational efficiency. Thus, this study aimed to compare the efficiency of classification algorithms in e-commerce service chatbots. The researcher used machine learning techniques with KNN and Random Forest algorithms in this case. To assess the feasibility of the application, the chatbot results will be tested using the confusion matrix method to assess accuracy. From this study, it was obtained that the KNN method and calculating word weight using TF-IDF produces an accuracy value of 71.4%, thus confirming its feasibility.

Enhancing Abstractive Multi-Document Summarization with Bert2Bert Model for Indonesian Language

Fri, 31 Jan 2025 00:00:00 +0700

This study investigates the effectiveness of the proposed Bert2Bert and Bert2Bert+Xtreme models in improving abstract multi-document summarization for Indonesians. This research uses the transformer model to develop the proposed Bert2Bert and Bert2Bert+Xtreme models. This research uses the Liputan6 data set which contains news data along with summary references for 10 years from October 2000 to October 2010 and is commonly used in many automatic text summarization research. The model evaluation results using ROUGE-1, ROUGE-2, ROUGE-L, and BERTScore show that the proposed model has a slight improvement over previous research models, with Bert2Bert being better than Bert2Bert+Xtreme. Despite the challenges posed by limited reference summaries for Indonesian documents, content-based analysis using readability metrics, including FKGL, GFI, and Dwiyanto Djoko Pranowo, revealed that the summaries produced by Bert2Bert and Bert2Bert+Xtreme are at a moderate readability level, meaning they are suitable for mature readers and aligns with the news portal's target audience.

Android Malware Threats: A Strengthened Reverse Engineering Approach to Forensic Analysis

Ridho Surya Kusuma, M Dirga Purnomo Putra — Fri, 31 Jan 2025 00:00:00 +0700

The widespread adoption of Android devices has rendered them a primary target for malware attacks, resulting in substantial financial losses and significant breaches of user privacy. Malware can exploit system vulnerabilities to execute unauthorized premium SMS transactions, exfiltrate sensitive data, and install additional malicious applications. Conventional detection methodologies, such as static and dynamic analysis, often prove inadequate in identifying deeply embedded malicious behaviors. This study introduces a systematic reverse engineering framework for analysing suspicious Android applications. In contrast to traditional approaches, the proposed methodology consists of six distinct stages: Initialization, decompilation, static analysis, code reversing, behavioral analysis, and reporting. This structured process facilitates a comprehensive examination of an application's internal mechanisms, enabling the identification of concealed malware functionalities. The findings of this study demonstrate that the proposed method attains an overall effectiveness of 84.3%, surpassing conventional static and dynamic analysis techniques. Furthermore, this research generates a detailed list of files containing specific malware indicators, thereby enhancing future malware detection and prevention systems. These results underscore the efficacy of reverse engineering as a critical tool for understanding and mitigating sophisticated Android malware threats.