Jurnal Sistemasi (OJS FTIK - UNISI, Fakultas Teknik dan Ilmu Komputer Universitas Islam Indragiri)
Not a member yet
946 research outputs found
Sort by
Optimization of CNN Activation Functions using Xception for South Sulawesi Batik Classification
Get PDFBatik motifs from South Sulawesi such as the Pinisi boat, Lontara script, Tongkonan house and Toraja combinations embody rich cultural narratives but are difficult to identify automatically. Automatic classification supports cultural preservation and education and empowers tourism and digital heritage applications. This study improves the performance of convolutional neural networks for South Sulawesi batik classification by optimizing activation functions within the Xception architecture which exploits depthwise separable convolutions for efficient and detailed feature extraction. A balanced dataset of 1400 labeled images in four classes was divided into eighty percent for training, ten percent for validation and ten percent for testing. Images were resized to 224 by 224 pixels, converted to grayscale and augmented through zoom, flip and rotation. With identical hyperparameters including a learning rate of 0.001, a batch size of 64 and training for 100 epochs using the Adam optimizer, ReLU, ELU, Leaky ReLU and Swish activation functions were compared. Evaluation metrics comprised accuracy, precision, recall, F1 score and cross entropy loss. ELU achieved the highest test accuracy of 98.57 percent, precision of 0.9864, recall of 0.9857 and F1 score of 0.9857, outperforming ReLU and Leaky ReLU with 97.86 percent accuracy and Swish with 97.14 percent accuracy. The results demonstrate that selecting an optimal activation function substantially enhances convolutional neural network classification of complex batik patterns. The findings offer practical guidance for development of resource aware batik identification systems in support of cultural digitization and education initiatives
Developing Web-Based Patient Reservation and Data Management System using Rapid Application Design
Get PDFA private infant and toddler care clinic in the South Tangerang area, Omah Bayi, faces problems in managing reservations and patient data. During this time, reservations were recorded in a book and scheduling was done verbally. This causes inefficiency and increases the possibility of data recording errors. To solve this problem, a web-based reservation system was developed using the rapid application development (RAD) method. This method involves requirements analysis, design, implementation, and system testing, with the support of technologies such as PHP and MySQL. The designed system effectively shortens the turnaround time of the reservation process and patient data management. Test results show that reservation recording time is reduced from 10 minutes to two minutes (80% faster), and monthly report preparation time is reduced from two hours to 10 minutes (92% faster). This research aimed to create a digital solution that will not only improve the clinic's operational efficiency and prevent data duplication but will also make services more accessible to patients. It is expected that the implementation of this system will be the first step towards the clinic's digital transformation and will have a positive impact on the development of healthcare services in the future
Payment Status Classification Invoice Bank Using Logistic Regression and Random Forest
Get PDFPayment management is an essential aspect of a bank’s financial operations, particularly in ensuring the smooth execution of procurement transactions for goods and services. The invoice, as an official document, plays a role in determining whether a transaction can be processed promptly or experiences a delay. Despite its central role, empirical research exploring the factors influencing invoice payment status remains limited, especially within the context of banking institutions. This study aims to analyze the factors that affect invoice payment status based on company type, procurement type, and invoice value. The methods employed include logistic regression and random forest to compare the classification performance of both approaches. The analysis reveals that procurement type and invoice value significantly influence payment status, with invoice value emerging as the most dominant variable based on the smallest p-value. In the random forest model, invoice value also ranks highest in terms of variable importance. In terms of accuracy, the random forest model outperforms logistic regression, achieving an accuracy of 94.47% compared to 59.30%. Although both methods yield similar precision (approximately 97%), random forest demonstrates a substantially higher recall (97.41%) and F1-score, whereas logistic regression records a recall of only 69.19%. These findings suggest that random forest is a more effective method for predicting payment status and holds strong potential for supporting data-driven decision-making in bank payment management system
Public Sentiment Analysis on TikTok about Tapera Policy using Random Forest Classifier
Get PDFAt the beginning of 2024, the Tapera policy proposed by the government sparked widespread public debate, resulting in both pros and cons. To improve the quality of public services, it is crucial for the government to evaluate policies to align with the needs and expectations of the community. This study aims to analyze public sentiment on the social media platform TikTok regarding the Tapera policy. Comment data was collected from several TikTok videos discussing the Tapera policy with high view counts. These videos received various responses in the form of comments, expressing positive, neutral, and negative sentiments about Tapera. A total of 5,036 comments were successfully scraped. The Random Forest Classifier was used for sentiment classification. This method was chosen for its ability to maintain high predictive accuracy, minimize overfitting, and perform effectively in classification tasks. The study results showed that negative sentiment dominated TikTok users' opinions, accounting for 82%, followed by neutral sentiment at 10% and positive sentiment at 8%. Many expressed disapproval for various reasons, including concerns about potential corruption, the ineffectiveness of contributions due to inflation, and the policy being burdensome amid a sluggish economy. Neutral sentiment was dominated by questions related to Tapera, such as the amount of Tapera deductions and whether participation is mandatory for those who already own a house. Positive sentiments expressed support for the Tapera policy and willingness to pay the contributions. However, the proportion of supporters of this program was significantly smaller than those opposing it. The training results of the classification model using the Random Forest Classifier achieved an accuracy of 89%. The highest F1-score for detecting negative sentiment was 94%, while the F1-score for detecting neutral sentiment was 17% and for positive sentiment, it was 32%. This disparity is due to the dataset composition being dominated by negative sentiment. The proportion of sentiment significantly influences the training of the classification model. A balanced proportion for each sentiment would enable the model to better learn and recognize the words frequently associated with each sentiment
The Moving Average Algorithm for Forecasting Palm Oil Fresh Fruit Bunch (FFB) Prices
Get PDFFresh Fruit Bunches (FFB) of oil palm are the main raw material in the production of Crude Palm Oil (CPO). Efforts to forecast FFB prices are crucial to help mitigate the negative impacts of price fluctuations, which are influenced by various external and internal factors. The problem addressed in this study is the fluctuation in FFB prices, which presents a major challenge for both farmers and stakeholders in the palm oil industry. The method used in this research is a quantitative approach with descriptive methods, applying the Moving Average algorithm. The results of the study show that the Moving Average algorithm is used to forecast FFB prices by calculating the average price over a specific time period. The accuracy of the forecast is evaluated using several metrics: first, the Root Mean Squared Error (RMSE) with a value of 137.19, indicating moderate forecasting error; second, the Mean Absolute Percentage Error (MAPE) with a value of 6.09%, indicating good accuracy; and third, the Mean Absolute Error (MAE) with a value of 117.0, indicating relatively small errors. The accuracy system is derived by comparing the forecasts with actual data and calculating metrics such as RMSE, MAPE, and MAE. These metrics help assess the accuracy of the FFB price forecasting model and allow for adjustments to improve forecasting results
Application of Double Exponential Smoothing Method for Forecasting Laptop Sales
Get PDFPT Indo Bismar is a retail company focused on laptop sales. The company experiences fluctuations in laptop sales each month, which impacts inventory management as it becomes challenging to predict demand accurately. Consequently, PT Indo Bismar faces financial losses due to unsold laptops. To address this issue, a sales forecasting system has been designed to optimize inventory management more effectively and efficiently.This study applies the double exponential smoothing method to forecast laptop sales and uses the Mean Absolute Percentage Error (MAPE) to measure forecasting accuracy. The double exponential smoothing method was tested through a trial-and-error approach. This process produced varying alpha and beta values for different laptop brands and models. It involved repeated iterations to test each combination until the optimal values that yielded the best forecasting accuracy were identified. After obtaining the MAPE results through the trial-and-error approach, the average system MAPE was calculated to evaluate the overall accuracy of the system, resulting in 16.58%. This indicates that the sales forecasting system demonstrates good accuracy, as the error rate falls within the range of 10% to 20%. Therefore, the use of the double exponential smoothing method can assist PT Indo Bismar in managing inventory and making strategic decisions for future laptop sale
Analysis of the E-Performance System and Competence on Employee Performance with Work Motivation as a Mediator
Get PDFThis study aims to assess the impact of the e-performance system and human resource (HR) competencies on employee performance, with work motivation as an intervening variable at the Regional Secretariat Office of Indragiri Hilir Regency. Using Structural Equation Modeling (SEM) with Partial Least Squares (PLS) 3.0, data were collected through a questionnaire distributed to 177 respondents. The analysis revealed several key findings. First, the e-performance system positively and significantly influences work motivation. Similarly, HR competencies have a positive and significant effect on work motivation. Additionally, the e-performance system directly impacts employee performance in a positive and significant way. However, the relationship between HR competencies and employee performance was found to be positive but not significant. Work motivation, on the other hand, has a positive and significant influence on employee performance. Furthermore, work motivation mediates the relationship between the e-performance system and employee performance, as well as between HR competencies and employee performance, both with positive and significant effects. These findings highlight the critical role of motivation in enhancing employee performance through e-performance systems and HR competencies
Application of Word2Vec and LSTM Models in Sentiment Analysis of Mobile Legends User Reviews
Get PDFSentiment analysis has become an important aspect of understanding user opinions regarding a product or service, including in the gaming industry. This study implements a combination of Word2Vec and Long Short-Term Memory (LSTM) models to analyze the sentiment of user reviews for the game Mobile Legends, obtained from the Google Play Store. The dataset used comprises 100,000 reviews that have undergone preprocessing stages such as text cleaning, tokenization, and stopword removal. The Word2Vec model is employed to represent the text in the form of numerical vectors, while LSTM is used to predict the sentiment of the reviews. Evaluation results indicate that this model achieves an accuracy of 87.88%, demonstrating the effectiveness of this method in classifying user sentiment. Further analysis reveals that the majority of user reviews are positive, with words such as "good," "exciting," and "awesome" frequently appearing in the word cloud. This research can provide insights for game developers in understanding user opinions and serve as a reference for the application of deep learning in sentiment analysis within the gaming industry
Literature Review: A Comparative Study of Waste Classification using Deep Learning Algorithms
Get PDFWaste type classification remains a daily challenge in modern waste management. Proper waste classification contributes significantly to environmental protection and enhances the efficiency of the recycling process. Unfortunately, manual waste classification is rarely performed by individuals, resulting in mixed waste that is difficult to separate into recyclable and non-recyclable categories. This leads to increased waste accumulation, which becomes harder to process over time. Therefore, automating this procedure using computer vision is of critical importance. This study adopts a Systematic Literature Review (SLR) methodology to analyze existing research conducted by previous scholars. The main objectives are to identify the most appropriate algorithms for waste type classification, determine the most suitable model architectures, and examine the correlation between dataset size, number of classes, and classification accuracy. The results of the literature review show that the Convolutional Neural Network (CNN) algorithm is widely used and considered highly effective for computer vision tasks. Among the best-performing models are: A standard CNN architecture achieving 100% accuracy with 150 data points and 3 classes, CNN with ResNet50 model achieving 99.41% accuracy on 2,527 data points and 6 classes, A combination of ResNet, k-Nearest Neighbors (kNN), and Neighborhood Component Analysis (NCA) achieving 99.35% accuracy on 13,089 data points and 1,672 classes, CNN with CapSA ECOC + ANN model reaching 99.01% accuracy on 1,515 data points and 12 classes. These findings indicate that numerous prior studies have successfully developed high-accuracy models for waste classification, which can serve as a solid foundation for building computer vision systems to automate the waste sorting process
Social Media Adoption in the Marketing Strategy of Radiant Buket SME: A UTAUT Model Approach
Get PDFThe advancement of digital technology has encouraged SMEs to adopt social media as part of their marketing strategies to enhance customer engagement and expand market reach. This study aims to analyze the factors influencing social media adoption by Radiant Buket SME using the Unified Theory of Acceptance and Use of Technology (UTAUT) model. The key variables examined include performance expectancy (PE), effort expectancy (EE), social influence (SI), and facilitating conditions (FC), which are hypothesized to affect behavioral intention (BI) and actual use behavior (UB) of social media. Radiant Buket’s marketing strategy involves the use of platforms such as Instagram and TikTok for product promotion, direct customer interaction, and the dissemination of information and testimonials through visual content. The findings show that PE and SI have a positive influence on BI, indicating that perceived usefulness and social encouragement drive the intention to adopt social media. Although EE has a negative effect, it still significantly influences BI, suggesting that users are willing to adopt social media despite perceiving it as requiring more effort. Additionally, FC positively affects UB, and BI significantly impacts UB, indicating that strong intention indeed leads to actual usage behavior. These findings offer practical insights for Radiant Buket in strengthening its digital marketing strategy, particularly through customer-oriented promotional content, enhanced two-way interaction on social media, and the optimization of support resources such as staff training and digital infrastructure. Theoretically, this study enriches the literature on technology adoption in the context of Indonesian SMEs, particularly in applying the UTAUT model within digital marketing. It also highlights the importance of a data-driven approach in developing sustainable and effective social media marketing strategies