The field of machine learning is rapidly advancing, and selecting the most suitable algorithm for predictive tasks remains a critical challenge. This study evaluates the predictive performance of three prominent machine learning algorithms: Deep Belief Networks (DBNs), Fuzzy ARTMAP, and Support Vector Machines (SVMs). Experiments on pattern recognition using image data from construction sites showed that DBNs achieved the highest predictive accuracy. In this study, experiments were conducted on a pattern recognition problem using image data from construction sites. The experimental results demonstrated that DBNs exhibited the highest predictive accuracy with the data used in this study. Algorithms such as DBNs, Fuzzy ARTMAP, and SVMs are representative models of machine learning methods, and their predictive power can vary depending on the type of data and the problem context. Therefore, future research should incorporate extended analyses with more diverse datasets and problem domains. Nonetheless, the findings of this study provide valuable guidelines for selecting appropriate algorithms for practical problem-solving and offer practical insights for practitioners aiming to optimize predictive accuracy across various machine learning applications. Bidang pembelajaran mesin berkembang pesat, dan memilih algoritma yang paling sesuai untuk tugas-tugas prediktif masih merupakan tantangan penting. Studi ini memberikan evaluasi komprehensif terhadap kinerja prediktif dari tiga algoritma pembelajaran mesin terkemuka: Deep Belief Networks (DBNs), Fuzzy Adaptive Resonance Theory Mapping (FuzzyARTMAP), dan Support Vector Machines (SVMs). Dalam penelitian ini, percobaan dilakukan pada masalah pengenalan pola menggunakan data gambar dari lokasi konstruksi. Hasil eksperimen menunjukkan bahwa DBN menunjukkan akurasi prediksi tertinggi dibandingkan data yang digunakan dalam penelitian ini. Algoritma seperti DBN, FuzzyARTMAP, dan SVM merupakan model representatif dari metode pembelajaran mesin, dan kekuatan prediksinya dapat bervariasi bergantung pada jenis data dan konteks masalah. Oleh karena itu, penelitian di masa depan harus menggabungkan analisis yang diperluas dengan kumpulan data dan domain masalah yang lebih beragam. Meskipun demikian, temuan penelitian ini memberikan pedoman berharga dalam memilih algoritma yang tepat untuk pemecahan masalah praktis dan menawarkan wawasan praktis bagi para praktisi yang ingin mengoptimalkan akurasi prediksi di berbagai aplikasi pembelajaran mesin.

Although input vector properties are the main determinants of neural network performance, most studies use a simple linear scaling model to normalize the input vector without considering input vector normalization.Because single normalization maps only a single property of the input data, the current normalization approach may not be effective for improving prediction accuracy.In addition, a linear type of normalization is just a linear mapping that mirrors raw data properties.On the other hand, non-linear normalization can reflect non-linearly by emphasizing certain properties.This study aims to propose a new and efficient normalization method that can provide better prediction performance by learning machines.For this purpose, various regularization methods such as linear and nonlinear mapping models and a combination of the two models were considered.The research methodology is as follows.a) Based on theoretical studies on linear and nonlinear models of regularization methods, a method that combines two different normalization methods is proposed.b) Data are prepared for experiments in neural network models.c) Experiments are conducted with the neural network model on two network topologies with three different normalization methods and 30 trials with different initial weights, and the results are obtained.d) The prediction performance is compared, and the best normalization method for traffic scene analysis is suggested.Experimental studies have shown that the prediction performance of the input vector by the nonlinear normalization method is 0.6-2.4% better than the prediction performance of the input vector by the linear method.More importantly, the predictive performance of the combination of the two normalization methods is 2.4-7.2%better than the linear method and 2.8-4.0%better than the non-linear method.The results of this study revealed that the combined use of two different normalization methods provides better performance in machine learning compared to individual normalization methods, such as linear and nonlinear normalization on input vectors.

This study aims to explore major neural network models - Stacked Denoising Autoencoder (SDAE), Deep Belief Network (DBN), Backpropagation - that have recently garnered attention and propose the most suitable and reliable artificial neural network model for real-time road traffic information collection. In this study, to enhance the reliability of experimental results, numerous experiments were conducted under identical conditions (such as parameter values and network configuration) by setting different initial values for the weight vector. The results of the experiments were statistically validated to draw conclusions. The research results showed that the SDAE model exhibited the most superior performance, while the accuracy of the DBN was somewhat lower compared to the SDAE model. On the other hand, the Backpropagation model demonstrated a relatively low predictive accuracy compared to both models, particularly showing a significant influence of the initial values

Artificial Neural Networks (ANNs) have been applied to many problems of traffic engineering and transportation planning domains. Diverse types of ANNs models, such as Backpropagation, Convolutional Neural Network (CNN) are currently exploring in many research areas, while Backpropagation is kwon as one of the most widely used neural network model among of them. Recently, CNN has received much attention in various fields because of its excellent object recognition and classification performances. The purpose of this study is to compare the prediction performance between the two models, general backpropagation and CNN model, in traffic scene analysis. The results of this study show that like backpropagation model, the initial weight vector of CNN model developed for this study plays a significant role its prediction performance in terms of accuracy. However, the overall experimental results show that the CNN model can provide better predictive performance than the Backpropagation model.

An intelligent transportation system (ITS) generally contains an automatic traffic video-surveillance system as a primary subsystem.Such subsystem incorporates the capabilities of neural networks for the efficient and effective recognition and classification of complicated spatial and temporal patterns in real-world traffic scenarios.Notably, the properties of input vectors are the key factors in determining the performance of neural networks.These properties are governed by the method used to normalize these vectors; a simple linear scaling model is widely employed for normalizing input vectors.This study proposes the use of a nonlinear normalization model for input vector normalization.The proposed technique is subsequently applied to neural networks to resolve classification problems encountered when analyzing real-world traffic image data.The experimental results show that the proposed model can produce higher prediction accuracy, when compared to the existing linear-based approach models.This model has the potential to improve the performance in traffic machine vision applications.

Air quality issues are now a major concern worldwide, and vehicle emission models are of great importance. Currently, most vehicle emission models are based on statistical analysis and it is necessary to develop new models to achieve better estimation accuracy. In addition, Artificial Neural Networks have proven to be superior to statistical models in various disciplines.In this study, Extreme Learning Machines was proposed to replace the existing statistical model, and the superiority of the proposed model was demonstrated by comparing the prediction performance with the backpropagation neural network model. Experimental results show that Extreme Learning Machines can outperform normal backpropagation models.

A comprehensive ITS (Intelligent Transport Systems) needs a dynamic parking information system giving real-time information about the number of parking space available in each car park. Computer vision has been potentially an important element in the foundation of integrated traffic surveillance and control systems. The objective of this study is to propose a robust and reliable method for detecting a vehicle and vacancy simultaneously in a car park using image processing techniques and a neural network model which is claimed to be more powerful than many expert systems.In this study, a neural network model is used for recognizing different patterns that come from complicated real scenes of a car park. The experiments have been done with the gray values of pixels on different image sizes and different gray levels in the image in order to find good input vectors to the network and pre-processing for reducing computing time. The results show that the method for vacancy and vehicle detection proposed in the paper is efficient for the noise, and therefore has the potential for parking data collection. In addition, the proposed method could be the framework for numerous traffic applications, such as congestion and incident detection, and vehicle classification with resulting improvements in performance to the method currently used.

There has been a wealth of literatures for the method of estimating an accurate origin-destination (O-D) trip matrix information from observed link counts since O-D trip matrix is one of the key information for a variety of transportation planning and analysis studies. A new type of applications, Machine learning (ML), recently opens a new door to transportation study including O-D trip matrix (ODME) domain so that many agencies and researchers are trying to incorporate both traditional ODME and emerging techniques including ML and Deep Learning (DL). Unfortunately, several difficulties such as limited computing power and data resources make such changes remain as unsolved. In this study, we offer a novel framework, an open-data-driven hybrid origin-destination trip matrix estimation model integrating with machine learning techniques (Open-DOMEiM). Several public resources including annual average daily traffic (AADT), public GIS map, American community survey (ACS), and average travel time data in normal and congestion conditions obtained from Google Map API, were used. Also, Several ML techniques including t-distributed stochastic neighbor embedding (t-SNE), Backpropagation (BP), Random forests (RF), and Stacked Autoencoder (SAE) algorithms which is one of the DL algorithms, were applied for this study. The t-SNE model was used as dimensional reduction algorithm to reduce computational burdens for other ML algorithms. An urban area of St. Louis, Missouri, USA was selected for the model comparisons along with one mathematical ODME model, Cho (2008). The results show that RF in conjunction with t-SNE and SAE outperformed other models for three scenarios in this study.

Traffic incidents have a serious negative impact on safety and traffic flow, and fast accurate automatic incident detection on freeways is a major theme in transportation engineering.Therefore, various types of AID (Automated Incident Detection) algorithms have been proposed for more accurate and rapid incident detection, and Artificial Neural Network models have provided significantly improved performance in terms of detection and false alarm rates.Recently, Deep Neural Networks (DNNs) has received much attention due to its excellent performance and was also used for automatic incident detection on highways.However, in learning algorithms such as Backpropagation and SVMs(Support Vector Machines), the prediction performance is known to be highly depend on the input vector characteristics.The purpose of this study is to examine whether the input detection performance of DNNs differs according to the normalization method of the input vector and to verify how sensitive it is to the method.Furthermore, the best way to normalize the input vector of the DNNs model has been proposed in order to obtain the best performance in terms of DR (Detection Rate) and FAR (False Alarm Rate) in AID (Automatic Incident Detection).

Automatic video-based vehicle detection is one of the main research topics in Intelligent Transportation Systems ( ITS) and is a key element for automatic traffic surveillance systems. Support Vector Machines ( SVMs) have been increasingly applied to an automatic video-based vehicle detection and a road scene analysis because of their remarkable performance in prediction accuracy. The property of input data for learning on SVMs determines the predictive performance. It is important task to choose the best input vectors in order to improve the predictive performance. It is normal to use a single property of input vectors in the application of learning models. However, the composition of different input vectors may affect predictive performance and a new input vector will be created by combining two raw data. In this paper, two information sources of edge information and pixel gray value have been combined to detect vehicle in road scene images and see how the fused data affect the predictive performance in SVMs. The experimental results of this study show that the fused data may provide better performance in predictive accuracy than a raw input data. Moreover, the results show that SVMs could provide much better performance than the Backpropagation model which is the best known neural network.

Support Vector Machines (SVMs) are being increasingly applied to automatic video-based object detection and traffic scene analysis because of their remarkable performance in real-time pattern recognition. However, the performance of SVMs is highly dependent on the property of input vectors. Studies have shown that the predictive performance of ANNs could be improved by preprocessing the original raw data. The purpose of this study was to investigate the predictive performance of SVMs with preprocessing methods, such as image size reduction and gray level compression. Experimental results from SVMs were compared with those from ANNs. This study showed that preprocessing techniques, which have been widely used to improve performance in various learning models, tend to undermine the predictive accuracy in SVMs learning model.

Automatic traffic scene analysis which has been used for real-time on-road vehicle detection system is essential to many areas of ITS (Intelligent Transport Systems). In order to improve the detection time and accuracy of detection performance, various image processing techniques have been used for real-time vehicle detection. Moreover, Neural Networks have been increasingly and successfully applied to many problems for ITS research topics. Support Vector Machines (SVMs) are currently another efficient approach to vehicle detection because of their remarkable performance. In this research, two different models, Backpropagation which is the best-known neural network model and SVMs have been studied to compare their performance in predictive accuracy, through experiment with real world image data of traffic scenes. Experimental results show that SVMs can provide higher performance in terms of predictive performance than the well-known Backpropagation neural network model.

The cost of delays caused by incidents is significant and a traffic management center needs to quickly detect and remove them from the freeway to reduce the impact of an incident. A study on quick and efficient automatic incident detection has been an important field of transportation research and many algorithms based on loop detector data have been developed for automatic incident detection on freeways. However, several simple and familiar algorithms, such as the California algorithms, the McMaster algorithm, and the Minnesota algorithm, have had limited success in their overall performance in terms of detection rate, false alarm rate, and mean time to detect an incident. Recently, detection algorithms based on the neural network models are known as the one of the most popular and efficient approaches for real-time automatic incident detection. Moreover, many researchers have shown that the neural network models were much more efficient than various other previous models. However, various types of neural network models and learning mechanisms have been developed and it is an important task to choose the best model in order to achieve the best performance in automatic incident detection. In a previous study, support vector machine (SVM), which is based on statistical learning theory, has been shown to be more efficient than Backpropagation algorithm, easily the most useful and popular neural network. In this paper, FuzzyARTMAP, which is a combination of fuzzy logic and adaptive resonance theory, has been used for automatic incident detection. Experiments have been carried out using real world freeway incident and incident-free data collected on freeway segments located in Seoul, Korea. Acomparative study with three models, Backpropagation, SVM, and FuzzyARTMAP, has been conducted to assess the incident detection performance in terms of detection rate and false alarm rate. Experimental results in this study showed that FuzzyARTMAP might provide better performance than the Backpropagation and SVM models for automatic incident detection.

Numerical problems that are usually ignored in the dynamic analysis of hydraulic control valves are described, and an analysis of the effects of such problems on the numerical modeling is provided. Previous studies have ignored the effects of changes in the flow coefficient in the orifice, the solenoid force along the spool movement in the valve and an ascending tendency of pressure during reach to the steady state. To eliminate these problems, it was studied a method to substantiate the non-linearity of the pressure loss caused by passing between the orifice and port as well as that caused by interaction with the solenoid. Moreover, the movement of the spool and spring is expressed using the time-delay-element (TDE). The proposed numerical model has been used in the Bond graphs method of a hydraulic control valve and the simulation results have been shown to be accurate. It is known that differences between simulated and experimental results can have a considerable impact on the function of actual systems. The contribution of each parameter is measured separately for the transient state and steady state. Analysis standard observed the first peak value, pressure increase to the steady state and the settling time in the response results.

In this paper, we propose a new method to refine edge information of the up-sampled depth image. After a low-resolution depth image is enhanced by its corresponding color image using joint bilateral filtering, the edge extracted from the depth image is interpolated. Then, the bi-linearly interpolated depth image is refined using the edge map. Experiments show that the proposed method outperforms the M-3DVC method in terms of the quality of the depth data and synthesized virtual view.

A numerical problem that has easily ignored in the dynamic analysis of hydraulic control valves is described, and an analysis of the effects of such problems on numerical modeling is provided. Previous studies have ignored the effects of changes in the flow coefficient in the orifice, the solenoid force along the spool movement in the valve and an ascending tendency of pressure during reach to the steady state. Thus, simulation results obtained in earlier studies have had numerical value errors. To eliminate these problems, this study employed a method to substantiate the nonlinearity of the pressure loss caused by passing between the orifice and port as well as that caused by interaction with the solenoid. Moreover, the movement of the spool and spring expressed using the time-delay-element (TDE). The proposed numerical model has been used in the Bond graph method of a hydraulic control valve and the simulation results have been shown to be accurate. It is known that differences between simulated and experimental results can have a considerable impact on the function of actual systems. The contribution of three factor mentioned in this paper was observed in TDE, Flow coefficient, solenoid order.

Research on neural networks has grown significantly over the past decade, with valuable contributions made from many different academic disciplines. While there are currently many different types of neural network models, Back-propagation is the most popular neural network model. However, the input vectors in the Back-propagation neural network model usually need to be normalized and the normalization methods affect the prediction accuracy. In this study, a new method is proposed in which an additional feed-forward process was included in the Back propagation model and a sigmoid activation function was modified, in order to overcome the input vector normalization problem. The experimental results showed that the proposed approach might produce a better training and prediction accuracy than the most current common approach using input vector normalization and that it has the potential to improve performance in machine vision applications.

Artificial neural networks, which hold considerable potential for recognising and classifying spatial and temporal patterns, have been used as an efficient method for automatic traffic surveillance, which is an important research topic of intelligent transport systems. An important element in the performance of the neural networks is the composition of the input and output vectors, as well as the network architecture. However, there has been little research on the performance in relation to the attributes of the input and output vectors. In this research, various input vector properties were applied to the backpropagation model, which is the most popular neural network model, to see how the general performance would be affected by the different types of input vector. Experiments were performed with the inclusion of various grey levels, image sizes, edge detection images and combinations of edge and pixel grey information as the input vectors. The experimental results showed that the network performance, in terms of computing cost for training and prediction accuracy, was highly dependent on the characteristics of the input vectors. Two combined input vectors, the grey scale pixels and edge detection image, produced better prediction performance than either the grey values on the pixel or edges alone.

The issue of air quality is now a major concern around the world and the vehicle emissions model is very important. Most of the current vehicle emission models are multiple regression techniques. In this study, a neural network-based model has been proposed to achieve better estimation accuracy.The estimation performance of two models, the proposed neural network-based model and a general regression model, has been compared using mean absolute error (MAE). A comparative study between two models to estimate vehicle emissions, the proposed neural network-based model and a general regression model, has been conducted to assess the estimation performance of the proposed model in terms of mean absolute percentage error. Experimental results in this study revealed that the neural network model performed better as it was able to decrease the error for emission estimation comparing with the multiple regression models.More importantly, in this study a lookup table (LUT) method has been proposed to overcome the black-box problem, which is a disadvantage of the neural network models. It could be useful for any other researches to estimate emissions without developing and training the neural network model which can be a time-consuming task.