Papers by Tanasanee Phienthrakul
Information and Communication Technology for Competitive Strategies (ICTCS 2020), 2021
2023 IEEE Symposium on Industrial Electronics & Applications (ISIEA)
Proceedings of the 2nd International Conference on Intelligent Systems, Metaheuristics & Swarm Intelligence
Sentiment classification on textual reviews refers to classifying textual reviews based on whethe... more Sentiment classification on textual reviews refers to classifying textual reviews based on whether they are positive or negative. This research focuses on classifying movie reviews, and is benchmarked on the IMDB dataset, which consists of long movie reviews, using accuracy as the evaluation metric. In sentiment classification, each document must be mapped to a fixed length vector. Document embedding models map each document to a dense, low-dimensional vector in continuous vector space. This research proposes to train document embedding using cosine similarity instead of dot product. Experiments on the IMDB dataset show that accuracy is improved when using cosine similarity compared to using dot product, while using feature combination with Naïve-Bayes weighted bag of n-grams achieves a new state of the art accuracy of 97.4%.
2018 International Electrical Engineering Congress (iEECON)
Dengue fever is a tropical disease caused by dengue virus. This virus is transmitted by mosquitoe... more Dengue fever is a tropical disease caused by dengue virus. This virus is transmitted by mosquitoes. Dengue can also be transmitted via infected blood products and through organ donation. There are many infected people in each year. Thus, if the number of patients can be predicted, vaccines for dengue fever will be able to prepare for serving all patients in that area. This research proposed a two-step prediction method that combines time series forecasting analysis and supervised learning techniques to predict the number of dengue fever patient cases. In generally, the result of prediction from only the number of patient cases is not good enough for application; accuracy and confidence are low. In this research, the environmental factors are considered and predicted. Then, these factors are used for predicting the number patient cases. The experimental results show that the proposed two-step prediction technique is a good choice for dengue fever prediction. The accuracy is better than the simple prediction technique.
2021 13th International Conference on Knowledge and Smart Technology (KST), 2021
An electronic nose has been applied in many areas such as the food industry, the environmental ar... more An electronic nose has been applied in many areas such as the food industry, the environmental area, and this technology can be used to detect some explosives. Many classification machine learning techniques are applied for creating the model which manipulates data into a defined group to be used for customer grouping, marketing, anomaly detection, and medical analysis. The purpose of this research is to find a suitable classification technique to be applied in an electronic nose to imitate the ability of sniffer dogs to detect the chemical substances. This research compares the accuracy of eight different classification techniques, which are Logistic Regression, Support Vector Machine (SVM), Decision Tree, Random Forest (RF), Adaptive Boosting, K-Nearest Neighbors, Gaussian Naive Bayes, and Multilayer Perceptron in both binary and multi-class gas sensor array open source datasets. Experimental results show the top algorithms are RF, and SVM models, which give average score as 99.66 and 98.93, respectively.
2019 14th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP), 2019
This paper proposed the query-by-example word spotting model for handwritten documents with image... more This paper proposed the query-by-example word spotting model for handwritten documents with image fuzzification. Fuzzy size of word images was used to size of problem. The number of classes in each set were decreased, which made it is easy to choose parameters. The Pyramid of Histogram of Oriented Gradients (PHOG) feature and Support Vector Machine (SVM) were employed to use in the model. IAM handwritten database was used for evaluating the model. The result demonstrates that the micro precision of model with image fuzzification and without image fuzzification were 35.11% and 23.54% respectively. However, the accuracies of the models were 35.11% and 40.14% respectively. Thus, the image fuzzification can be used for reducing of type one error with slightly accuracy loss.
Lecture Notes in Networks and Systems, 2021
Kernel functions are used in support vector machines (SVMs) to compute inner product in a higher ... more Kernel functions are used in support vector machines (SVMs) to compute inner product in a higher dimensional feature space. The performance of classification or approximation depends on the chosen kernel function. There are some popular kernel functions such as linear, polynomial, and radial basis function (RBF) kernels. However, these common kernel functions may not be sufficient for the complex or large problems. This research proposes to improve the performance of SVM by using the non-negative linear combination of these common kernel functions. The obtained kernel functions are more flexible and allow better discrimination or approximation in the feature space. Then, the evolutionary strategies (ESs) are used for adjusting the parameters of SVM and the proposed kernel functions. In order to avoid the overfitting problem, the objective function in the evolutionary process is carefully designed. Training error, subset cross-validation, the bound of generalization error, and the st...
2016 International Computer Science and Engineering Conference (ICSEC), 2016
Many new devices come out with the idea of making more comfortable life. One of these ideas is an... more Many new devices come out with the idea of making more comfortable life. One of these ideas is an Myo Armband which is a wireless device for interacting with other devices such as smartphone or computer to use as a mouse, keyboard, or even game controller by using Bluetooth technology and electromyography (EMG) sensor. In order to communicate with computer, user has to pose his/her hand and arm to matched the specified patterns for sending a command to the computer. However, the pose detection from itself is failed to detect the real pose from user even if it is synced following the guideline from its developer. This may cause some error in the application that depended on accuracy of the pose detection. Concept of this paper presents another way to detect the poses by using decision tree to solve the problem by sending EMG data sets from all 8 EMG sensors around it for deciding the pose from user. Rock-paper-scissors game is created to test the concept of Myo Armband training. The experiment results show that the proposed technique can use as a pose detection algorithm in around 78% correction.
2018 10th International Conference on Knowledge and Smart Technology (KST), 2018
Many new devices come out with the idea of making more comfortable life. Myo armband is a wireles... more Many new devices come out with the idea of making more comfortable life. Myo armband is a wireless device for interacting with computer using electromyography (EMG) sensor. To communicate with the computer, the poses of hand and arm are matched with the command to control like a mouse click. Although the standard Myo can be used to communicate with computer, some poses cannot be detected or their results may be wrong. In this paper, the machine learning techniques will be applied to detect the hand gestures or poses. Double-tap, fist, spread finger, wavein, and wave-out are 5 basic poses. These basic poses and rest will be trained and tested. The experimental results show that RBF network yields the acceptable results when it is compared to the results of many techniques.
2018 International Conference on Information Technology (InCIT), 2018
Technology and information are used in everywhere. Technology has massive impact on the society, ... more Technology and information are used in everywhere. Technology has massive impact on the society, but its stages are still early. Many applications were implemented for logistic and manufacturing process to personal usage. This paper focuses on the application of IoT on a personal level, which is home automation. The curtain are driven by a direct-current motor. The lights are operated according to the light sensor. Both of them are controlled by a touch sensitive smart mirror or voice activation. With the help of an online database management program and ESP8266 wireless devices, the communication between the microcontrollers was done successfully. System activity can be tracked via NETPIE. Results showed that curtain and lights were able to be controlled remotely via touch control and voice activation. Record of results are kept for future machine learning integration. In practice, benefits of IoT were demonstrated in order to bring down the barriers and create a pathway to mainstream adaptation of IoT smart devices.
Proceedings of the 9th International Conference on Machine Learning and Computing, 2017
Signature is a kind of biometric identification that is widely used in contracts, agreements, and... more Signature is a kind of biometric identification that is widely used in contracts, agreements, and other legal documents. However, the signature may be imitated by deceivers. In order to verify the signature, specialists are required and they may use a lot of time to inspect the suspect signatures. This paper proposes to verify the signatures using machine learning techniques, such as neural network, decision tree, decision table, and Naive Bayes. Feature extraction is an important part in learning process. Concept of chain code is introduced and this concept will be used for extracting some features from signature images. These features and geometric features are used to train and test in order to verify the actual signatures. The experimental results show that the proposed features can improve the accuracy of signature verification.
This paper presents a method for palm and palm’s lines detection based on image processing techni... more This paper presents a method for palm and palm’s lines detection based on image processing techniques. An application of the proposed method is illustrated in the automatic palmistry system. Both hardware and software are created and tested. The system can detect palm and three main lines, i.e., life line, heart line, and brain line. Line’s position, line’s length, and line’s curvature are used for palmistry prediction. These three lines will be compared to the lines in the line pattern archives by using the nearest neighbor method. The experimental results show that this system can detect palm and palm’s lines and this system yields the suitable results on many examples. Furthermore, the concept of this system can be applied to identification and authentication in security approaches or in the embedded system fields.
2015 3rd International Symposium on Computational and Business Intelligence (ISCBI), 2015
Feature subset selection is an important problem in machine learning and data mining. If the suit... more Feature subset selection is an important problem in machine learning and data mining. If the suitable features are selected, the results of classification or prediction will be more accurate, while if the unsuitable features are used, the results may have no meaningful. This paper presents a method for feature subset selection that uses the ensemble technique to increase the efficiency of feature selection. Association rule mining is introduced to select the high relationship features. Bagging concept is applied to increase the confidence of selection. The experimental results show the efficiency of the proposed method that outperforms the efficiency of simple association feature subset selection.
2018 10th International Conference on Knowledge and Smart Technology (KST), 2018
The objective of this research is to develop an Ultraviolet monitoring, alert, and prediction sys... more The objective of this research is to develop an Ultraviolet monitoring, alert, and prediction system (UV-MAPs). This system can measure UV-index, show heatmap of UV-index, alert on risk or effect from UV-index around user, predict UV-index, and show UV intensity on mobile application. UV-indexes are recorded between 6 a.m. and 6 p.m. around Mahidol University Salaya campus. The related data, i.e., UV intensity, temperature, humidity, and location are recorded from sensors. Then, the heatmap of UV-index is created. The future UV-indexes are predicted by linear regression. The outputs of prediction are sent to mobile application. The experimental results show that the predicted UV-index is very similar to the index that was predicted by a famous website. Heatmap can accurately illustrate UV-index around Mahidol University, Salaya campus. The data from many sensors are recorded and they are used for training and testing. Moreover, mobile application can alert when users are in the risk...
This paper proposes the methods for solving the traveling salesman problems using clustering tech... more This paper proposes the methods for solving the traveling salesman problems using clustering techniques and evolutionary methods. Gaussian mixer model and K-means clustering are two clustering techniques that are considered in this paper. The traveling salesman problems are clustered in order to group the nearest nodes in the problems. Then, the evolutionary methods are applied to each cluster. The results of genetic algorithm and ant colony optimization are compared. In the last steps, a cluster connection method is proposed to find the optimal path between any two clusters. These methods are implemented and tested on the benchmark datasets. The results are compared in terms of the average minimum tour length and the average computational time. These results show that the clustering techniques are able to improve the efficiency of evolutionary methods on traveling salesman problems. Moreover, the proposed methods can be applied to other problems.
Kernel functions are used in support vector regression (SVR) to compute the inner product in a hi... more Kernel functions are used in support vector regression (SVR) to compute the inner product in a higher dimensional feature space. The performance of approximation depends on the chosen kernels. The radial basis function (RBF) kernel is a Mercer's kernel that has been widely used in many problems. However, it still has the restriction in some complex problems. In order to obtain a more flexible kernel function, the multi-scale RBF kernels are combined by nonnegative weighting linear combination. This proposed kernel is proved to be a Mercer's kernel. Then, the evolutionary strategy (ES) is applied for adjusting the parameters of SVR and kernel function. Moreover, subsets cross-validation is used for evaluating these parameters. The optimum values of these parameters are searched by (5+10)-ES. The experimental results show the ability of the proposed method that outperforms the statistical techniques.
Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop, 2019
In document-level sentiment classification, each document must be mapped to a fixed length vector... more In document-level sentiment classification, each document must be mapped to a fixed length vector. Document embedding models map each document to a dense, lowdimensional vector in continuous vector space. This paper proposes training document embeddings using cosine similarity instead of dot product. Experiments on the IMDB dataset show that accuracy is improved when using cosine similarity compared to using dot product, while using feature combination with Naïve Bayes weighted bag of n-grams achieves a new state of the art accuracy of 97.42%.
2016 Fifth ICT International Student Project Conference (ICT-ISPC), 2016
The amounts of data are growing continuously. The data can make a benefit for the organization if... more The amounts of data are growing continuously. The data can make a benefit for the organization if they have the right plan to collect and analyze the data. In this paper, we examine data on research citation information. Many authors create interesting articles for propagating information to other researchers. The relationship network of researchers is also growing continuously. Learning network of researchers is necessary to find who has the most influence on others in the network. The researchers do not only have many papers but also they have co-authors who may stay in different communities. Adaptive information from various topics will make original papers. The combination of knowledge among researchers from various communities is a good way to create interesting papers. The aim of this article is to present a new measurement for author evaluation by using clustering coefficient and weighted degree centrality. The result will be used to rank researchers in order and analyze properties of the top 5 researchers. The ranking result can be comparable to the popular usage method, h-index. Hence, the new measurement for author evaluation using social network analysis measurement is a good way for author ranking.
Uploads
Papers by Tanasanee Phienthrakul