KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 17

Some of the metrics are blocked by your
consent settings
Vehicle Travel Time Estimation in Transportation Network Using Random Forest and Neural Network
(2025-01-01)
Nakano, Shuya
;
Panichpapiboon, Sooksan
;
Kulla, Elis
Several technologies in Intelligent Transportation Systems (ITS), such as automatic driving, electric vehicles, vehicular communications, are reshaping the way we travel and use the transportation system. Automatization of ITS mainly consists of traffic management, traffic light control, optimal route selection and so on. In these automatic applications, accurate estimation of vehicle travel time is essential to make efficient decisions. This study uses synthetic traffic data generated by Simulation of Urban MObility (SUMO) to evaluate machine learning models, like Random Forest and Neural Network, for travel time estimation. A variety of traffic-related features were collected, and three feature scenarios were tested. Results show that the Random Forest model outperforms both the neural network and the baseline method based on numerical estimation, highlighting the benefit of feature-rich approaches.
Some of the metrics are blocked by your
consent settings
A Novel Classification Model Based on Hybrid K-Means and Neural Network for Classification Problems
(2024-09-01)
Chenghu, Cui
;
Thammano, Arit
We propose a new classification model—a new classification model for clustering overlapping problems based on K-Means and neural networks. K-means clustering algorithm belongs to unsupervised learning. It is a classic algorithm for solving clustering problems. Since this algorithm calculates its categories based on distance, the results tend to converge to the local optimal solution and have poor boundary clustering properties. The K-Means classification algorithm defines clusters by the distance between the cluster center value and the target object, and the optimal result is obtained through continuous iteration. Therefore, clustering results are overlapped, and there are often outliers that do not belong to the current cluster, resulting in unsatisfactory clustering results. Our model offers a new method to segment non-ideal data in overlapping regions. Since clustering algorithms cannot effectively identify and classify this part of the data, we split this part of the data and train it using a neural network. The results are then integrated into the clustered data. In the experiment, the k-fold cross-validation method ensures the model stability of the results. We used the accuracy to evaluate the quality of the model, and we used standard deviation and mean deviation to detect clustering results. Five sets of experimental data from the cross-experiment show that compared with the K-Means classification model, the accuracy of our model is effectively improved.
Some of the metrics are blocked by your
consent settings
An Improved Neural Manufacturing Corporate Credit Rating Model Based on LSTM
(2024-01-01)
Zhang, Rui
;
Chen, Binbin
;
Sakdanuphab, Rachsak
This paper proposes an improved neural manufacturing corporate credit rating model based on Multi-head Self-attention (MSA) mechanism and Long Short-Term Memory (LSTM) network. The proposed model leverages MSA to simulate the marketdynamics and generate dynamic weights for each indicator based on the financial dataof all manufacturing companies. Meanwhile, LSTM is utilized to extract sequential features from long-term financial and operational data to capture the long-term financial status and reduce the risk of deviation. The experimental results show that the proposedmodel provides more objective and reliable credit ratings for manufacturing companies.In the comparison experiment with the baseline model, it was proven that the model proposed in this paper outperforms other baseline models. In the comparison experiment with SMAGRU, it was proven that the proposed model has better predictionability than SMAGRU on both datasets, and it also demonstrates that the GRU simplifies the internal computation of LSTM. The ablation experiment verified the feasibility of the two modules of the proposed model separately, which further proved the effectiveness of the proposed model.
Some of the metrics are blocked by your
consent settings
Instrumental Receiver Bias Estimation for Ionospheric Total Electron Content by Neural Network Model
(2023-10-01)
Thu, Phyo C.
;
Supnithi, Pornchai
;
Budtho, Jirapoom
;
Saekow, Apitep
;
Sopon, Thanomsak
Total Electron Content (TEC) is one of the most important parameters in the study of the ionosphere, especially for determining ionospheric disturbances. The TEC levels are typically estimated from dual-frequency GPS observation data. Since the measured TEC contains discrepancies such as satellite and receiver biases, they need to be removed to obtain more accurate TEC values. In this work, we estimate the receiver bias using a neural network technique. Based on the exhaustive evaluation, we design a neural network (NN) model with two-hidden layers, and it is trained with datasets from three GNSS observation stations in Thailand. The prediction from the proposed neural network deviates from the baseline reference using the minimum standard deviation method with significantly faster computational time. The trained NN model is also tested for estimating the receiver bias values at other untrained stations in Thailand.
Some of the metrics are blocked by your
consent settings
Reference Signal Received Power Prediction Using Convolutional Neural Network with Residual Loss
(2023-01-01)
Ngenjaroendee, Thearrawit
;
Phakphisut, Watid
;
Wijitpornchai, Thongchai
;
Areeprayoonkij, Poonlarp
;
Jaruvitayakovit, Tanun
In this paper, LTE measurement reports collected from user equipments are used to generate the residual loss, which can represent the loss value of each grid. The residual loss and geospatial data are used in the learning process of convolutional neural network (CNN). We also use the site configuration and three-dimensional antenna pattern. Thus, the neural network and convolutional neural network are proposed to construct deep learning to predict the reference signal received power (RSRP) in Bangkok, Thailand. The results show that residual loss can improve the efficiency of prediction.
Some of the metrics are blocked by your
consent settings
Comment Usefulness Classification on Youtube using Artificial Neural Networks
(2023-01-01)
Takhom, Akkharawoot
;
Chirawat, Pimmada
;
Boonkwan, Prachya
Social media represents a vast and constantly evolving data resource, utilized across various domains, including business. However, discerning valuable data for business purposes demands significant analysis and labor, leading to potential errors. To address this, we propose employing deep learning models for classifying useful data, focusing on a case study of quality comments on YouTube in the Thai language. Specifically, we experiment with four sequence-To-sequence models: Recurrent Neural (RNN), Long Short-Term Memory (LSTM), bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU). The experimental results reveal that the Bi-LSTM model exhibits the most promising performance, achieving an accuracy of 94%. Furthermore, Bi-LSTM demonstrates remarkable precision of 0.94, recall of 0.95, and an F1-score of 0.95, underscoring its proficiency in the precise classification of quality comments.
Some of the metrics are blocked by your
consent settings
AI-Enhanced Predictive Maintenance in Manufacturing Processes
(2022-01-01)
Netisopakul, Ponrudee
;
Phumee, Nawarat
This research aims to apply artificial intelligence technology to a manufacturing industry, specifically, to forecast temperature and insulation values of motors from the CNC machine. Dataset from motor sensors are collected and forecasting models are trained using four deep learning models, namely, multilayer perceptron (MLP), long-short term memory (LSTM), LSTM autoencoder, and bidirectional LSTM (Bi-LSTM). Models are evaluated by measuring the deviation of forecasting values from the real values. Two measures, root mean square error (RMSE) and mean absolute error (MAE), are used to assess model's performance. Experiments are conducted and found that the Bi-LSTM yielded the lowest RMSE and MAE numbers, hence, the best model to be selected. Further development has been implemented by integrating Bi-LSTM and genetic algorithm (GA) in order to optimize the model performance. Instead of searching the huge hyperparameter space of the neural network, the integrate GA-LSTM model using RMSE as a fitness function to reduce the search space and obtain the optimal or near optimal hyperparameters. The empirically best model is found which yields a lower RMSE value of 0.041 comparing to 0.18 when not optimized.
Some of the metrics are blocked by your
consent settings
Optimal neuro-fuzzy equalizers for nonlinear channels of the perpendicular magnetic recording system
(2021-01-01)
Wongsathan, Rati
;
Supnithi, Pornchai
Nonlinear distortions caused by partial erasure and nonlinear transition shifts interacting with inter-symbol interference, are a major hindrance to data storage systems, since they degrade detector performance. This work aims to design and optimize the neuro-fuzzy equalizer (NFE) using the multi-objective genetic algorithm (MOGA) to detect nonlinear high-density magnetic recording (MR) channels. Through the GA-assisted back-propagation algorithm and least mean square optimization, the complexity in terms of decision rules is reduced by 25% and significantly provides 65% lower signal processing computation. When applied to the perpendicular (MR) system, the proposed NFE outperforms existing equalizers such as the neural network-based equalizer, fuzzy logic equalizer, and conventional NFE for the Volterra and jitter media noise channels using 1-3 dB and 1.5-3.5 dB signalto-noise ratio gains at the bit-error-rate of 10−4, respectively. Furthermore, compared to the other models, the NFE provides a more effective output mean square error performance for retrieving the original bit data.
Some of the metrics are blocked by your
consent settings
Holy Basil Curl Leaf Disease Classification using Edge Detection and Machine Learning
(2020-02-14)
Tangtisanon, Pikulkaew
;
Kornrapat, Suttipong
Holy basil (Ocimum basilicum L.) is one of the most vital economic crops that has a significant impact on export earnings. However, the holy basil prices could be dropped due to a curl leaf disease caused by pests. Several previous studies focused on plant leaf disease detection based on the leaf color. Unfortunately, the leaf curl disease sometimes changes a shape of the leaf not the color so it cannot be detected with those schemes. We proposed a novel approach aims to automatically detect a curling leaf on holy basil. This paper presents a Neural Network (NN) model and Logistic Regression (LR) model to automatically detect a curling leaf on holy basil. To be able to detect the infected one not by colors but by its shape, we have applied edge detection algorithms which are Canny and Sobel model. To speed up processing time, images were resized and converted to grayscale before passing them to machine learning models. Moreover, NN and LR were modified with mini-batch technique in order to increase the speed of the processing time. The dataset contains 600 images of holy basil leaves with 300 images of healthy leaves and 300 images of infected leaves. The experimental results indicate that the proposed method effectively detects the curling leaves on holy basil.
Some of the metrics are blocked by your
consent settings
Automatic classification of pararubber trees in Thailand from LANDSAT-8 images using neural networks method
(2019-07-01)
Supunyachotsakul, Chisaphat
;
Anan, Thanwarat
;
Suksangpanya, Nobphadon
;
Nakya, Suvalak
Classifying features from satellite images has been a time-consuming manual process which requires lots of manpower. This work exploits deep convolutional decoder encoder neural network, to develop an algorithm that can automatically classify the extents of the Pararubber tree growing areas from the LANDSAT-8 images. The classification resulted from this approach was verified. In conclusion, the classification accuracy achieved is at 86.90% with Cohen's kappa at 73.80% which is considered satisfactory.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results