Journal of Science Ladkrabang

Clause-Level Subjective Classification for Thai Article Using Bidirectional Long Short-Term Memory

2023-06-08T02:32:31+07:00

Sentence subjective classification is one of the crucial steps in analyzing opinions from such data as articles or online media which the volume has increased greatly. Extracted opinions from sentences can be used as information to produce or improve products. This research presented a method to create a model for classifying opinion at the clause level in Thai language articles using a Bidirectional Long Short-Term Memory (BiLSTM) deep learning model. This model is widely used to deal with sequential data. Moreover, the FastText model was used to convert words into numerical vectors. Our research experimented by creating models from texts in multi-domain and measuring the accuracy of the classification using the LST20 dataset. This dataset contains 44,423 pre-segmented clauses, including Part of Speech and Named Entity annotations, which are used as features for model learning. The evaluation of model performance used 5-fold cross-validation. We found that the BiLSTM model using 200 neurons in the Long Short-Term Memory unit with word and Part of Speech as features is the best model. It achieved precision of 62.562%, recall of 51.151%, accuracy score of 79.407%, and F1-score of 56.284%.

Discriminant Analysis for Defect Reduction in Plastic Printing Process of Beverage Closures

2023-08-08T10:52:36+07:00

This research aimed to create the discriminant analysis equation for a classification group of good and bad products, in order to reduce the defects of the plastic printing process by means of the discriminant analysis. According to the company's case study, there was a problem with the beverage closures getting damaged because of the trademark ink peeling off. It was crucial for the brand of the goods and for attracting consumers to buy the product. A total of 180 sample data sets were collected. There were the following three predicted variables: gas pressure (x₁), compressed air pressure (x₂), and conveyor speed (x₃). The dependent variables were separated into two categories: good components and defects. In this research, the discriminant analysis equation could be determined as , under a 94.4% accuracy rate. The suitable gas pressure, compressed air pressure, and conveyor speed were 82 mbar, 7 psi, and 1000 rpm, respectively. These results were replicated experimentally, which proved that every beverage closure was good.

An Application of Liu Logistic Regression Using the Bootstrap Method for Sex Classification from Sacrum of Thai Population

2023-04-12T09:38:29+07:00

The purpose of this study aimed to classify the gender from Sacrum of Thai population by Liu Logistic Regression via using Bootstrap method. The results come from the experiment on 78 observations, those contain 46 men and 32 women. There are 13 covariates. The response variable is sex. Those covariates are X₁ (the mean of Midventral straight length), X₂ ( the mean of Midventral curved length), X₃ (the mean of Ventral straight breadth), X₄ (the mean of Transverse diameter base), X₅ (the mean of Transverse base), X₆ (the mean of AP diameter body s 1), X₇ (Breadth of alae), X₈ (the mean of Sacral index (%)), X₉ (the mean of Longitudinal curvature index (%)), X₁₀ (the mean of Corporobasal index (%)), X₁₁ (the mean of S1 index), X₁₂ (the mean of Alea index), and X₁₃ (the mean of Sacral base index). The iteration is 500. A parameter estimation is shrinkage D6. Data is separated into 3 cases, those cases are 60:40, 70:30 and 80:20. The criteria is to compare the performance of model is F1 Score from the calculation based on confusion matrix. This research found that the 70:30 is the best case from the model on every point of Sacrum. The value yield the highest F1 score when compared to other cases is 97.74%.

Exploring the Association Rules of Road Traffic Accidents

2023-05-25T11:49:05+07:00

In this research, the objective is to extract patterns of road traffic accidents in Thailand using association rules and the FP-growth algorithm was employed to mine frequent patterns. The road accident data in Thailand used in this study was collected from Information and Communication Technology Center, Office of the Permanent Secretary of the Ministry of Transport between January 1, 2019 and September 30, 2022, with a total of 74,231 instances and 10 attributes included time, month, provinces, weather, types of cars, accident causes, accident form, road characteristics, road types, and results of accidents. The association analysis results on Thailand road traffic accidents showed that the most common patterns of road traffic accidents were rear-end collisions, rollovers, and falls onto the road. The speed limit violation led to the collision. Accidents are more likely to occur when the road is straight and no slope, and clear weather. Furthermore, the most common types of vehicles engaged are motorcycles, while the roads of the Ministry of Highway were the scene of the most collisions.

Comparing K-Mean Clustering Methods of DNA in Brain Tumors for High-Dimensional Data

2023-03-12T23:57:50+07:00

This study aims to compare the performance of clustering DNA of brain tumor patients of k-means three methods, namely the Hartigan-Wong, Forgy, and MacQueen methods. The independent variables are DNA as 989 genes, and the dependent variable is the level of a brain tumor in 43 patients. In this case, the number of the independent variable is larger than the number of patients or called the high-dimensional data. The experiment is conducted by random DNA samples of 200, 400, 600, and 800 genes and fixed 5, 10, 15, 20, 25, and 30 groups by 1,000 replications. Comparing clustering performance is the mean data differences between the groups' criteria. The results of k-means clustering methods find that the Hartigan-Wong method has the best performance for all situations. However, the Hartigan-Wong method shows the most significant difference in data between groups compared to Forgy and MacQueen methods. The number of independent variables has not affected clustering performance.

The Development of a Semantic-based Image Retrieval Model by Pre-training Neural Network

2023-08-18T21:17:19+07:00

This research aims to develop a semantic-based image retrieval model applying the Contrastive Language-Image Pre-training (CLIP) model. Evaluation of image retrieval performance with precision, recall and f-measure, it was found that image search results with the query by global labels condition and the query by high level concepts of the images condition had a very good level of precision, the model can efficiently retrieve images from the content. However, image retrieval results with the query by qualitative semantic concepts of the image condition, despite having a good level of precision. But the results are far from the user's expectations because the semantic of image is interpreted by experience on human perception principles. In addition, also, the semantic of image is difficult to evaluate whether they are correct or not. The output from this research can resolve the semantic gap problem and support users by query within a natural language that attaches to the semantic of the image rather than the grammar of the language. This impact of results in a guideline for semantic information retrieval in the future.

Jujube Classification from Images Using Deep Learning Technique

2023-08-11T12:35:16+07:00

This research presents the extraction of jujubes from photographs using a deep learning technique identify jujubes from photographs using the Inception V3 algorithm. Data sets used in this research are separated into four groups: a,b,c, and d, each group contains 300 images. Test data contains 50 images and data sets used to validate model contains 40 images for training and test models. The experimental results show that Inception V3 has the highest accuracy of 97%, compared to the accuracy of the CNN model of 89% and MobileNet V2 of 86%. It can be used for jujube classification, which is beneficial to farmers. The research findings for the classification data with different quality. The accuracy of the jujube classification model can be further improved.

A Comparison of Forecasting Models for Carbon Dioxide Emission Quantity from Power Generation Sector

2023-04-12T09:34:22+07:00

The objective of this research is to compare forecasting models for carbon dioxide emission quantity from power generation sector using the data from Energy Policy and Planning Office, Ministry of Energy from January 2001 to December 2022 with a total of 264 months. The data was divided into 2 sets. The first set of data was collected from January 2001 to December 2021 with a total of 252 months for constructing 3 forecasting models: Box-Jenkins method, Winters’ Additive method, and Combined forecasting method. The second set of data was gathered from January to December 2022 with a total of 12 months for comparing the suitability of the forecasting models by considering the lowest root mean square error and mean absolute percentage error. The results indicated that Box-Jenkins method was the most appropriate method.

Production of Coenzyme Q10 and Carotenoid by Cultivation of Rhodopseudomonas sp. OS33-UV13-5 Using Raw Starch as Carbon Source

2023-12-04T10:03:25+07:00

Rhodopseudomonas sp. OS33-UV13-5 is the photosynthetic bacterial mutant strain obtained from a genetic improvement of Rhodopseudomonas sp. OS33 by UV irradiation. Rhodopseudomonas sp. OS33-UV13-5 showed the ability to digest raw starch, resulting in a problem-solving of the medium viscosity and a decrease in an energy consumption due to the direct degradation of raw starch particles. The result showed that Rhodopseudomonas sp. OS33-UV13-5 gave the highest growth at 11.82±0.935 U_770nm, carotenoid concentration of 15.67±1.137 mg/L, and coenzyme Q10 concentration of 288.79±20.132 µg/L in cells cultivated in medium containing 300 g/L of raw starch under light intensity of 1,500 lux. They were 5.65, 2.00, and 1.44 times, respectively, higher than those from cells cultivated in medium containing 10 g/L of raw starch under light intensity of 1,000 lux. Under optimal conditions, a promising hydrolysis of raw starch to glucose as well as glucose utilization provided the lower glucose concentration than 2.0 g/L during a cultivation which promoted not only bacterial growth but also a production of carotenoid and coenzyme Q10. Finally, the relationship between the growth and raw starch amylase activity was demonstrated. During the cultivation of Rhodopseudomonas sp. OS33-UV13-5, the glucose level was found at 0.12±0.08 to 1.18±0.15 g/L.

A New Sixth Order Iterative Method with Three Step for Solving Nonlinear Equation

2023-09-14T15:07:38+07:00

This paper presented a new three-step iterative method, free from second derivative for solving nonlinear equation. The idea was derived from the two third-order convergent iterative methods, namely Halley's method and Sharma's method. These were integrated to enhance the Classical Chebyshev method, which also possessed third-order convergence. Moreover, an analysis of the convergence order for this new iterative method was considered and demonstrated sixth-order convergence. Additionally, a numerical example was provided to show the efficiency of this new iterative method.