International Journal of Engineering and Computer Science Applications (IJECSA)

Thesis Topic Modeling Study: Latent Dirichlet Allocation (LDA) and Machine Learning Approach

2024-09-06T08:45:21+08:00

The thesis reports housed in the campus repository have yet to be analyzed to reveal valuable knowledge patterns. Analyzing trends in thesis research topics can facilitate the selection of research topics, aid in mapping research areas, and identify underexplored topics.Therefore, this research aims to model and classify thesis topics using Latent Dirichlet Allocation (LDA) and the Naïve Bayes and Support Vector Machine (SVM) methods. This study employs the LDA method for thesis topic modeling, while SVM and Naïve Bayes are used for classifying these topics. The research results show that LDA successfully modeled five of the most popular thesis topics, namely two related to computer networks, two on software engineering, and one on multimedia. For thesis topic classification, the SVM method demonstrated higher accuracy than Naïve Bayes, reaching 92.80% after the data was balanced using Synthetic Minority Oversampling Technique (SMOTE). The implication of this study is that the topic modeling approach using LDA is able to identify dominant thesis topics. In addition, the SVM classification results obtained better accuracy than Naïve Bayes in the thesis topic classification task.

Enhancing Mental Illness Predictions: Analyzing Trends Using Multiple Linear Regression and Neural Network Backpropagation

2024-09-07T13:10:11+08:00

The increasing number of mental health cases caused by various factors such as social changes, economic pressures, and technological advancements has made it difficult to accurately predict the number of cases, hindering prevention and early intervention efforts. Therefore, developing more accurate, data-driven predictive models is necessary to improve the effectiveness of prevention and intervention. This study aims to develop a predictive model for the number of mental health cases using Multiple Linear Regression and Neural Network Backpropagation methods. The study employs two predictive methods, Multiple Linear Regression and Neural Network Backpropagation to forecast future trends in the number of mental health cases. The findings reveal that the Neural Network Backpropagation method provides more accurate predictions than Multiple Linear Regression in forecasting mental health case trends. Specifically, the Neural Network Backpropagation method resulted in an MAE of 111.39 and a MAPE of 1.77%, while the Multiple Linear Regression method produced an MAE of 115.24 and a MAPE of 1.83%. Thus, the implication of this study is that the Neural Network Backpropagation method can be utilized to predict trends in the number of mental health cases due to its ability to provide highly accurate predictions.

Website-Based Expert System for Diagnosing Epilepsy in Children Using the Forward Chaining Method

2024-11-08T13:43:54+08:00

Information technology has been used in various sectors of life, including the health sector. One of them is the use of expert systems in diagnosing disease. Disease diagnosis carried out by an expert has weaknesses along with the expert's biological weaknesses. One technology that can be a solution is an expert system. This research aims to build a web-based expert system for diagnosing epilepsy in children, along with things that parents can do when treating epilepsy patients. The method used in this research is forward chaining, and system testing is carried out using the Black Box method. From the system design that has been created and tested, a web-based expert system application for diagnosing epilepsy in children has been produced. The black-box testing results show that all menus function well and as expected. The results of expert testing and user testing results obtained a final score of 3.8, which means the assessment is in the very suitable category. Apart from that, it will provide information and education to the public, in this case, the parents of epilepsy patients, regarding the type of epilepsy the child suffers from and how to treat it, which can be accessed anywhere and at any time.

Clustering Analysis of Umrah Pilgrim Data Based on the K-Medoid Method

2024-11-22T08:25:14+08:00

The Umrah pilgrimage is becoming increasingly popular among Indonesians, with millions of participants yearly. This trend creates a need for service providers to understand the characteristics of pilgrims to improve service quality, marketing strategies, and competitiveness. Analyzing data on pilgrims helps service providers develop more effective strategies and tailor packages to match their needs, ensuring competitiveness in a growing market. This study aims to clusters Umrah pilgrims based on age, gender, district, and chosen package using the K-Medoid clustering method. This research uses the K-Medoid method for the reason that it is more resistant to noise and outliers compared to other clustering methods. The most centrally located point in the data set is called a ”medoid,” which is an object in a cluster that has the lowest difference to all other objects in the cluster. The results of this study are that the K-Medoid method successfully grouped pilgrims into three clusters: Cluster 1 with 63 members, Cluster 2 with 25 members, and Cluster 3 with 25 members. The findings indicate that the Milad Mastour package is preferred by older pilgrims, primarily from Mataram and West Lombok. The Arbain package is favored by younger pilgrims from the same regions, while adult pilgrims mostly choose the Regular package. The implication of this research is that it can provide insights for service providers to design more specific programs that align with the profiles of pilgrims based on age and district.