MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer

Analyzing the Application of Optical Character Recognition: A Case Study in International Standard Book Number Detection

Mon, 03 Feb 2025 01:58:23 +0000

In the era of advanced education, assessing lecturer performance is crucial to maintaining educational quality. One aspect of this assessment involves evaluating the textbooks authored by lecturers. This study addresses the problem of efficiently detecting International Standard Book Numbers (ISBNs) within these textbooks using optical character recognition (OCR) as a potential solution. The objective is to determine the effectiveness of OCR, specifically the Tesseract platform, in facilitating ISBN detection to support lecturer performance assessments. The research method involves automated data collection and ISBN detection using Tesseract OCR on various sections of textbooks, including covers, tables of contents, and identity pages, across different file formats (JPG and PDF) and orientations. The study evaluates OCR performance concerning image quality, rotation, and file type. Results of this study indicate that Tesseract performs effectively on high-quality, low-noise JPG images, achieving an F1 score of 0.97 for JPG and 0.99 for PDF files. However, its performance decreases with rotated images and certain PDF conditions, highlighting specific limitations of OCR in ISBN detection. These findings suggest that OCR can be a valuable tool in enhancing lecturer performance assessments through efficient ISBN detection in textbooks.

Revealing Interaction Patterns in Concept Map Construction Using Deep Learning and Machine Learning Models

Tue, 18 Feb 2025 07:10:13 +0000

Concept maps are educational tools for organizing and representing knowledge, enhancing comprehension, and memory retention. In concept map construction, much knowledge can be utilized. Still, concept map construction is complex, involving actions that reflect a user’s thinking and problemsolving strategies. Traditional methods struggle to analyze large datasets and capture temporal dependencies in these actions. To address this, the study applies deep learning and machine learning techniques. This research aims to evaluate and compare the performance of Long Short-Term Memory (LSTM), K-Nearest Neighbors (K-NN), and Random Forest algorithms in predicting user actions and uncovering user interaction patterns in concept map construction. This research method collects and analyzes interaction logs data from concept map activities, using these three models for evaluation and comparison. The results of this research are that LSTM achieved the highest accuracy (83.91%) due to its capacity to model temporal dependencies. Random Forest accuracy (80.53%), excelling in structured data scenarios. K-NN offered the fastest performance due to its simplicity, though its reliance on distance-based metrics limited accuracy (70.53%). In conclusion, these findings underscore the practical considerations in selecting models for concept map applications; LSTM demonstrates effectiveness in predicting user actions and excels for temporal tasks, while Random Forest and K-NN offer more efficient alternatives in computational.

Identify the Condition of Corn Plants Using Gray Level Co-occurrence Matrix and Bacpropagation

Abd Mizwar A. Rahim, Theopilus Bayu Sasongko — Thu, 06 Mar 2025 03:02:30 +0000

This research aims to increase the accuracy of identifying the condition of corn plants based on leaf features using the GLCM and ANN Backpropagation methods. The GLCM method is used to extract features from corn leaf images, while Backpropagation ANN is used to classify the condition of corn plants based on these features. This classification was carried out using a dataset of corn leaves from four different conditions, namely healthy, leaf-spot, leaf-blight, and leaf-rust. Next, leaf features are extracted using the GLCM method. After that, data normalization was carried out, balancing the dataset, and training was carried out on the Backpropagation ANN model to classify the condition of the corn plants. After training the model, the next model evaluation is carried out using the confusion matrix method. The research results show that the method used can produce quite high accuracy when identifying the condition of corn plants, with an accuracy of 99%. This shows that the use of GLCM and ANN Backpropagation can be a good alternative in identifying the condition of corn plants. This research provides benefits in making it easier to accurately identify the condition of corn plants.

Determining Toddler’s Nutritional Status with Machine Learning Classification Analysis Approach

Wed, 05 Mar 2025 00:00:00 +0000

The nutritional status of toddlers is a common issue many countries face worldwide. Various facts indicate that malnutrition is a primary focus for many researchers. Several efforts have been made to address this problem, including developing analytical models for identification, classification, and prediction. This study aims to evaluate the nutritional status of children by utilizing a classification analysis approach using Machine Learning. This research aims to improve the accuracy of the classification system and facilitate better decision-making in stunted toddlers, which is a priority, especially in the health sector. The Machine Learning classification analysis process will later utilize the performance of the Naive Bayes algorithm, the Support Vector Machine algorithm, and the Multilayer Perceptron algorithm. ML performance can be optimized using gridsearchCV to produce optimal classification analysis patterns. The data set of this study uses 6812 toddler data sourced from the Health Center at the Tangerang Regency Health Office. Based on the research presented, Machine Learning performance in analyzing nutritional status classification provides maximum results. The results are reported based on a precision level with an accuracy of 88%. The results of this analysis can also present a classification of nutritional status based on knowledge. This study can contribute to and update the analysis model in determining nutritional status. The results of this study can also provide benefits in handling nutritional status problems that occur in children.

Combination Forward Chaining and Certainty Factor Methods for Selecting the Best Herbs to Support Independent Health

Thu, 06 Mar 2025 03:27:44 +0000

The use of herbal medicine as an alternative treatment is increasingly popular due to its natural benefits and cultural significance. However, a lack of public knowledge about the effectiveness, appropriate dosages, and processing methods of herbal remedies poses a significant barrier to their proper utilization. This knowledge gap often leads to suboptimal or even unsafe usage of herbal medicines. To address this issue, this study proposes an application-based system combining the Forward Chaining and Certainty Factor methods to provide personalized recommendations for the best herbal remedies supporting self-health management. The research aims to enhance accessibility to reliable information on herbal treatments while ensuring accurate and user-specific recommendations. By utilizing the Forward
Chaining and Certainty Factor method, this system identifies suitable herbal plants based on the type of disease, processing techniques, recommended dosages, and duration of treatment. Meanwhile, the Certainty Factor method calculates the level of certainty for each recommendation provided. The study’s results showed a validation rate of 90%, indicating that the combination of these two methods effectively bridges the gap between traditional herbal knowledge and modern health needs. This study concludes that the system offers a practical tool for individuals to select and use herbal treatments safely and effectively, promoting better health outcomes.

Comparison of Text Representation for Clustering Student Concept Maps

Wed, 05 Mar 2025 00:00:00 +0000

This research aims to address the critical challenge of selecting a text representation method that effectively captures students’ conceptual understanding for clustering purposes. Traditional methods, such as Term Frequency-Inverse Document Frequency (TF-IDF), often fail to capture semantic relationships, limiting their effectiveness in clustering complex datasets. This study compares TF-IDF with the advanced Bidirectional Encoder Representations from Transformers (BERT) to determine their suitability in clustering student concept maps for two learning topics: Databases and Cyber Security. The method used applies two clustering algorithms: K-Means and its improved variant, K-Means++, which enhances centroid initialization for better stability and clustering quality. The datasets consist of concept maps from 27 students for each topic, including 1,206 concepts and 616 propositions for Databases, as well as 2,564 concepts and 1,282 propositions for Cyber Security. Evaluation is conducted using two metrics Davies-Bouldin Index (DBI) and Silhouette Score, to assess the compactness and separability of the clusters. The result of this study is that BERT consistently outperforms TF-IDF, producing lower DBI values and higher Silhouette Scores across all clusters (k= 2 - k=10). Combining BERT with K-Means++ yields the most compact and well-separated clusters, while TF-IDF results in overlapping and less-defined clusters. The research concludes that BERT is a superior text representation method for clustering, offering significant advantages in capturing semantic context and enabling educators to identify student misconceptions and improve learning strategies.

Sentiment Study of ChatGPT on Twitter Data with Hybrid K-Means and LSTM

Dimas Afryzal Hanan, Ario Yudo Husodo, Regania Pasca Rassy — Thu, 06 Mar 2025 00:00:00 +0000

The rapid evolution of artificial intelligence (AI) has transformed the way people interact with technology, with ChatGPT emerging as a standout innovation in natural language processing (NLP). While it offers immense benefits, such as improving productivity and accessibility, it has also sparked debates about trust, transparency, and user experience. This makes understanding public sentiment about ChatGPT both timely and essential.This study explores user sentiments by combining K-Means clustering and Long Short-Term Memory (LSTM) models for analysis. The research utilized a dataset from Kaggle, which underwent extensive preprocessing, including text cleaning, tokenization, and lemmatization. Key features were extracted using TF-IDF and Word2Vec techniques, while clustering was refined with the Elbow Method and Silhouette Score. The data was grouped into three clusters focusing on ChatGPT’s functions, its developers, and user activities. Sentiment analysis using LSTM achieved an impressive accuracy of 98% after five training cycles. The findings highlight that negative sentiments, particularly around technical challenges and transparency, dominate user feedback, signaling areas for improvement. While positive sentiments exist, they remain overshadowed by critical perspectives. This study underscores the importance of enhancing user trust and experience while ensuring ethical and transparent AI development. The insights provided aim to guide developers and policymakers in creating AI technologies that are more user-focused and socially responsible. Future research should include multilingual and cross-platform data to paint a more comprehensive picture.

Analysis of Combination Machine Learning Classification with Feature Selection Technique for Lecturer Performance Analysis Model

Sat, 08 Mar 2025 05:24:53 +0000

Machine learning-based classification techniques are widely utilized for accurate analysis in various fields. This study focuses on assessing lecturer performance in higher education to enhance teaching standards and produce high-quality learning outcomes. Previous studies have employed multiparameter approaches, such as statistical correlation analysis, but these methods fail to achieve optimal accuracy and precision due to limited alignment with data characteristics. This research proposes a lecturer performance measurement model by evaluating three machine learning algorithms: k-Nearest Neighbors (k-NN), Decision Tree, and Na¨pve Bayes. The model integrates three feature selection techniques to improve classification performance: ANOVA, Information Gain, and Chi-Square. The study aims to enhance classification accuracy and assess the impact of feature selection techniques on performance metrics. A significant contribution of this research is introducing a dynamic feature selection approach tailored to data characteristics, which improves classification model performance. The methodology comprises three main stages: data loading and measurement of relevant parameters; data preprocessing, including filtering, cleaning, transformation, normalization, and feature selection; and performance evaluation using a machine learning-based classification approach. Experimental results demonstrate that the Decision Tree algorithm combined with Chi-Square feature selection achieved an accuracy of 0.887, precision of 0.903, recall of 0.887, and F1-score of 0.884. The proposed model
provides a reliable framework for evaluating lecturer performance and can be utilized to recognize and reward high-performing lecturers effectively.

The Mitigating Overfitting in Sentiment Analysis Insights from CNN-LSTM Hybrid Models

Mon, 10 Mar 2025 03:18:16 +0000

This study aims to improve sentiment analysis accuracy and address overfitting challenges in deep learning models by developing a hybrid model based on Convolutional Neural Networks and Long Short-Term Memory Networks. The research methodology involved multiple stages, starting with preprocessing a dataset of 5,456 rows. This process included removing duplicate data, empty entries, and neutral sentiments, resulting in 2,685 usable rows. To overcome data quantity limitations, data augmentation expanded the training dataset from 2,148 to 10,740 samples. Data transformation was carried out using tokenization, padding, and embedding techniques, leveraging Word2Vec and GloVe to produce numerical representations of textual data. The hybrid model demonstrated strong performance, achieving a training accuracy of 99.51%, validation accuracy of 99.25%, and testing accuracy of 87.34%, with a loss value of 0.56. Evaluation metrics showed precision, recall, and F1-Score values of 86%, 87%, and 86%, respectively. The hybrid model outperformed individual models, including Convolutional Neural Networks (70% accuracy) and Long Short-Term Memory Networks (81% accuracy). It also surpassed other hybrid models, such as the multiscale Convolutional Neural Network-Long Short-Term Memory Network, which achieved a maximum accuracy of 89.25%. The implications of this study demonstrate that the hybrid model based on Convolutional Neural Networks and Long Short-Term Memory Networks effectively improves sentiment analysis accuracy while reducing the risk of overfitting, particularly in small or imbalanced datasets. Future research is recommended to enhance data quality, adopt more advanced embedding techniques, and optimize model configurations to achieve better performance.

Optimization of Content Recommendation System Based on User Preferences Using Neural Collaborative Filtering

Lusiana Efrizoni, Junadhi Junadhi, Agustin Agustin — Tue, 11 Mar 2025 03:21:47 +0000

Recommender systems play a crucial role in enhancing user experience across various digital platforms by delivering relevant and personalized content. However, many recommender systems still face challenges in providing accurate recommendations, especially in cold-start situations and when user data is limited. This study aims to address these issues by optimizing content recommendation systems using Neural Collaborative Filtering (NCF), a deep learning-based approach capable of capturing non-linear relationships between users and items. We compare the performance of NCF with traditional methods such as Matrix Factorization (MF) and Content-Based Filtering (CBF) using the MovieLens-1M dataset. The research method employed is a quantitative approach that encompasses several stages, including preprocessing, model training, and evaluation using metrics such as Root Mean Squared Error (RMSE) and Precision@K. The results of this research are significant, demonstrating that NCF achieves the lowest RMSE of 0.870, outperforming MF with an RMSE of 0.950 and CBF with an RMSE of 1.020. Additionally, the Precision@K achieved by NCF is 0.73, indicating the model’s superior ability to provide more relevant recommendations compared to baseline methods. Hyperparameter tuning reveals that the optimal combination includes an embedding size of 16, three hidden layers, and a learning rate of 0.005. Despite its excellent performance, NCF still faces challenges in handling cold-start cases and requires significant computational resources. To address these challenges, integrating additional metadata and exploring regularization techniques such as dropout are recommended to enhance generalization. The implications of the findings from this study suggest that NCF can significantly improve prediction accuracy and recommendation relevance, thus having the potential for widespread application across various domains, such as e-commerce, streaming services, and education, to enhance user experience and the efficiency of recommendation systems. Further research is needed to explore innovative solutions to address cold-start challenges and reduce computational demands.

Evaluation Analysis of the Necessity of Stemming and Lemmatization in Text Classification

Sun, 16 Mar 2025 05:35:10 +0000

Stemming and lemmatization are text preprocessing methods that aim to convert words into their root and to the canonical or dictionary form. Some previous studies state that using stemming and lemmatization worsens the performance of text classification models. However, some other studies report the positive impact of using stemming and lemmatization in supporting the performance of text classification models. This study aims to analyze the impact of stemming and lemmatization in text classification work using the support vector machine method, in this case, devoted to English text datasets and Indonesian text datasets, and analyze when this method should be used. The analysis of the experimental results shows that the use of stemming will generally degrade the performance of the text classification model, especially on large and unbalanced datasets. The research process consisted of several stages: text preprocessing using stemming and lemmatization, feature extraction with Term Frequency-Inverse Document Frequency (TF-IDF), classification using SVM, and model evaluation with 4 experiment scenarios. Stemming performed the best computation time, completing in 4 hours, 51 minutes, and 41.3 seconds on the largest dataset. While lemmatization positively impacts classification performance on small datasets, achieving 91.075% accuracy results in the worst computation time, especially for large datasets, which take 5 hours, 10 minutes, and 25.2 seconds. The Experimental results also show that stemming from the Indonesian balanced dataset yields a better text classification model performance, reaching 82.080% accuracy.

Integration of Image Enhancement Technique with DenseNet201 Architecture for Identifying Grapevine Leaf Disease

Rudi Kurniawan, Lukman Sunardi — Thu, 13 Mar 2025 15:11:23 +0000

Early detection of grapevine leaf diseases is crucial for maintaining both the quality and quantity of grape production. Manual identification methods are often ineffective and prone to errors. This research aims to develop a precise and efficient method for classifying grapevine leaf diseases using Contrast Limited Adaptive Histogram Equalization (CLAHE) and the DenseNet201 Deep Convolutional Neural Network (DCNN) architecture. The research methodology involves collecting a dataset of grapevine leaf images affected by black measles, black rot, and leaf blight alongside healthy leaves. Following this, preprocessing is conducted using the CLAHE technique to enhance image quality. Then, the processed data is trained with DenseNet201. Evaluation results indicate that the proposed model achieves an overall accuracy of 99.61%, with high precision, recall, and F1-score values across all disease classes. Receiver Operating Characteristic (ROC) curve analysis shows an Area Under the Curve (AUC) of 1.00 for each class, reflecting excellent discriminatory ability. The loss and accuracy curves illustrate consistent model performance without signs of overfitting. Additionally, the confusion matrix confirms very low classification error rates. The developed model is effective and reliable for identifying grapevine leaf diseases. Future research will focus on enhancing the dataset by incorporating more data optimizing hyperparameters, and developing field applications for real-time use.

Novel Application of K-Means Algorithm for Unique Sentiment Clustering in 2024 Korean Movie Reviews on TikTok Platform

Baiq Rima Mozarita Erdiani, Aryo Yudo Husodo, Ida Bagus Ketut Widiartha — Mon, 17 Mar 2025 00:00:00 +0000

In recent years, social media has become one of the main factors influencing public perception of films. As a rapidly growing video-sharing platform, TikTok plays a crucial role in shaping audience opinions through comments, short reviews, and user discussions. This phenomenon is increasingly relevant in the Korean film industry, attracting global attention with its diverse genres and engaging narratives. However, a deep understanding of how audiences respond to films based on genre remains limited, especially in the dynamic context of social media. Therefore, this study aims to analyze audience sentiment toward Korean films released in 2024 on TikTok, focusing on sentiment distribution across four main genres: comedy, romance, action, and fun stories. The research methodology includes data collection through web crawling on TikTok, followed by text preprocessing and feature extraction using IndoBERT. Sentiment classification uses SentimentIntensityAnalyzer to categorize comments into positive, negative, or neutral. Since the dataset consists of unlabeled text, K-Means clustering is employed to identify sentiment groupings, with validation using principal component analysis to ensure cluster quality. The findings indicate that the romance and comedy genres are predominantly associated with neutral sentiment, reaching 89.6% and 87.4%, respectively. In contrast, the action genre exhibits higher sentiment polarization, with 14.9% positive and 24.7% negative sentiment. The fun story genre shows a more evenly distributed sentiment pattern. The main challenges include determining the optimal number of clusters and addressing imbalanced sentiment distribution across genres. This study provides valuable insights for filmmakers and marketers to understand audience reactions on social media better, enabling more targeted promotional strategies. Additionally, it contributes to the literature on sentiment analysis in the film industry, emphasizing the importance of genre-specific audience reception patterns for future research.

Predicting Gen Z’s Sentiments on Gorontalo’s CulturalWisdom Using Sentiment Analysis Models

Hermila A., Rahmat Taufik R. L Bau, Sitti Suhada, Abdulaziz Ahmed siyad — Mon, 17 Mar 2025 00:00:00 +0000

In the digital era, Generation Z faces both opportunities and challenges in understanding and preserving local cultural wisdom. It is essential to ensure that cultural values go beyond superficial consumption in cyberspace and remain deeply valued as integral aspects of identity and history. As technological advances and globalization continue influencing young people to preserve, local culture presents an increasing challenge. This study aims to determine Generation Z’s perceptions of Gorontalo’s local cultural wisdom, focusing on the Dikili and Meeraji traditions through a sentiment analysis approach. This research method uses the Naive Bayes algorithm to analyze positive, negative, and neutral sentiments derived from text data processed through a structured pre-processing stage. The findings reveal that Generation Z’s perceptions of the Dikili and Meeraji cultures are mostly positive, reflecting a strong appreciation and acceptance of these cultural values. However, the presence of negative sentiment highlights a critical view among some members of Generation Z, who consider certain aspects of these traditions less relevant or controversial in a modern context. In addition, neutral sentiment indicates a segment of young people who may need more exposure or information to form an informed opinion. The study concludes that while Dikili and Meeraji cultures still hold value among Generation Z, a more inclusive and adaptive approach to cultural preservation is needed. The findings offer valuable insights for strategies to preserve and develop local cultural heritage in the digital age

A Image Classification of Poisonous Plants Using the MobileNetV2 Convolutional Neural Network Model Method

Mon, 17 Mar 2025 00:00:00 +0000

Poisonous plants can be dangerous for many people, but some can be used as medicines or as pest killers. Some people, especially those in environments with a wide variety of plants, can take advantage of this poisonous plant. Lack of knowledge and information causes the use of this poisonous plant to be inappropriate. This research aims to develop software to classify images of poisonous plants using the Convolutional Neural Network method with the MobileNetV2 model and to compare the accuracy of classification results with various dataset configurations and varying parameters. The research method used is a Convolutional Neural Network, which has relatively high accuracy in classifying various digital images. The data used in this research consists of eight poisonous plants and several non-poisonous plants. The research results on 153 test data show that the accuracy value was 99.34%, precision was 99%, recall was 99%, and F1-Score was 99%. This research contributes to developing software that can quickly provide information and knowledge about poisonous plants, offering a high-accuracy solution for classifying poisonous plants using image data. Furthermore, implementing MobileNetV2 provides an efficient and lightweight model suitable for deployment on mobile devices, enhancing accessibility and usability in the field. The potential applications of this software extend beyond individual use, potentially benefiting agricultural, medical, and educational sectors. Future work will expand the dataset to include more plant species and refine the model to improve its robustness against diverse environmental conditions and image qualities.