Search

The goal of feature selection methods is to find the optimal feature subset by eliminating irrelevant or redundant information from the original feature space according to some evaluation criteria. In the literature, the Relief algorithm is a typical feature selection method, which is simple and easy to execute. However, the classification accuracy of the Relief algorithm is usually affected by the noise. In recent years, the Monte Carlo Tree Search (MCTS) technique has achieved great success in strategy selections of large-scale systems by building a tree and quickly focusing on the most valuable part of the search space. In this paper, with the benefit of MCTS, an MCTS-based feature selection approach is proposed to deal with the feature selection problem of high dimensional data, where the Relief algorithm is used as the evaluation function of the MCTS approach. The effectiveness of the proposed approach is demonstrated by experiments on some benchmark problems.

The physical health test of college students is a comprehensive assessment of their physical health status from the body shape, physical function, physical fitness and other aspects. Study of the physical health test using data mining technology has important practical significance for the reform of the physical education in universities. This paper presents a quantitative analysis approach of physical health test based on adaptive proportional fusion ensemble classifier for college students. It can be used to mine the importance of the sports items which affect the physical health test score for college students. In the experiments, the physical health test data is used to analyze the importance of different sports test items on the total score of physical health for male and female college students. The results have an important guiding significance for improving the physical health test score of college students, and enhancing the pertinence of the training of key sports items for physical education teachers.

The neuron classification problem is significant for understanding structure-function relationships in computational neuroscience. By solving neuronal classification problem, we can further understand the characteristic of neurons and the process of information transmission. In the training process of the neuron geometry classifier, selecting the appropriate neuron feature data is the premise of obtaining good classification results. First, a data set of 200 three-dimensional neurons was constructed, and there were 10 different types of neurons. Secondly, according to the difference of the extracted neuron characteristic parameters, it is divided into two categories: global features and distribution features. Finally, the machine learning methods such as Logistic Regression, K-Nearest Neighbor, Naive Bayes, Decision Tree, Random Forest, Extremely Randomized Trees and Extreme Gradient Boosting are used to classify neurons, and the effects of different features on neuron classification results are analyzed. The experimental results show: (1) For the K-Nearest Neighbor and Naive Bayes, the distribution feature of the neuron is more critical than the global feature, and even if the effect of the combination of the two will not be improved; (2) The ensemble classifier based on decision trees, such as Random Forests, Extremely Randomized Trees and Extreme Gradient Boosting have better accuracy and stability for neuron classification.

Since biological neurons have complex and diverse spatial geometric structures, classifying neurons according to their geometries has been an important issue. At present, the main approaches of neuron classification generally extract the structural features of neurons by the specified methods, and then use such features for classification. However, there are mainly two problems for these approaches. One is that there is no feature extraction standard of neuronal morphology, and the other is that a lot of useful information for neuron classification may be lost. Therefore, using the convolutional neural network model, this paper proposes a new three-dimensional (3D) neuronal morphology classification approach without geometric feature extraction. This approach first converts SWC format data of neuron into the scaled 3D neuronal graphics, and obtains the voxel data of neuronal morphology, and then uses convolutional neural network for training and classification. The experimental results show that this approach can classify the brain region cholinergic neurons of drosophila melanogaster, and obtains better classification accuracy.

The neuron classification problem is significant for understanding structure-function relationships in computational neuroscience. Advances in recent years have accelerated the speed of data collection, resulting in a large amount of data on the geometric, morphological, physiological, and molecular characteristics of neurons. These data encourage researchers to strive for automated neuron classification through powerful machine learning techniques. This paper extracts a statistical dataset of 43 geometrical features obtained from 116 human neurons, and proposes a neuronal morphology classification approach based on deep residual neural networks with feature scaling. The approach is applied to classify 18 types of human neurons and compares the accuracy of different number of residual block. Then, we also compare the accuracy between the proposed approach and other mainstream machine learning approaches, the classification accuracy of our approach is 100% in the training set and the testing set accuracy is 76.96%. The experimental results show that the deep residual neural network model has better classification accuracy for human neurons.

Safety and quality of travel on arterial networks tie closely to the performance of signalized intersections. Measures commonly used for intersection performance evaluations are control delay, queue length, and cycle failure. However, these variables are not directly available from typical configurations of traffic sensors designed for intersection signal control. Collecting vehicle control delay data manually for intersection performance measurement has been a task too time-consuming and labor-intensive to be practical. Video image processors (VIPs) have been increasingly deployed for intersection signal control in recent years. This study aims to use the extra detection capabilities of VIPs for performance monitoring at signalized intersections. Most VIPs can support up to 24 virtual loops, but normally less than half of the virtual loops are used. By properly configuring the spare virtual loops and analyzing the loop measurements, intersection performance can be monitored in real time. In this research...

Traffic congestion is getting worse and has resulted in increased travel delays and costs. In order to develop effective intelligent transportation systems (ITS) strategies to mitigate traffic congestion on freeways, a good understanding of its causes and impacts is vital but has not been achieved at a satisfactory level. Over the past several decades, deterministic queuing theory (DQT) has been widely used to evaluate freeway travel delays resulted from traffic congestion. However, several studies evaluated the accuracy of its delay estimates and claimed that the DQT method consistently underestimates vehicle delays. The reason for the underestimation, however, had not been clearly identified. This study aims at exploring the main cause of such underestimation problems and proposing a solution to fix it. Based on theoretical analysis and empirical justification, it was found the underestimation resulted primarily from the inappropriate estimates of the time offsets, that is, the travel times between the ...

Pedestrian and bicycle monitoring is quickly becoming an avid area of interest as information regarding pedestrian and bicycle flow is needed not only for developing competent access to particular urban corridors and trails, but also for system optimization scenarios, such as transit system operations and intersection controls. In this paper, the authors present a simple, yet effective method for tracking pedestrian and bicycle objects in a relatively large surveillance area, using ordinary uncalibrated video images. Object extraction is accomplished via background subtraction, while tracking is accomplished through an inherent characteristic cost function. Composite objects are used as a means of dealing with occlusions. The algorithm is implemented using Microsoft Visual C# and was tested on numerous scenes of varying complexity, resulting in an average count rate of 92.7% at the specified checkpoints.

Signal cycle failure (or overflow) is an interrupted traffic condition in which a number of queued vehicles are unable to depart due to insufficient capacity during a signal cycle. Cycle failure detection is essential for identifying signal control problems at intersections. However, typical traffic sensors do not have the capability of capturing cycle failures. In this article, we introduce an algorithm for traffic signal cycle failure detection using video image processing. A cycle failure for a particular movement occurs when at least one vehicle must wait through more than one red light to complete the intended movement. The proposed cycle failure algorithm was implemented using Microsoft Visual C#. The system was tested with field data at different locations and time periods. The test results show that the algorithm works favorably: the system captured all the cycle failures and generated only three false alarms, which is approximately 0.9% of the total cycles tested.

Traffic monitoring cameras are widely installed on streets and freeways in U.S. metropolitan areas. Video images captured from these video cameras can be used to extract many valuable traffic parameters through video image processing. A popular way to capture traffic data is to compare the current traffic images with the background image, which contains no vehicles or other moving objects, just background such as pavement. Once the moving vehicle images are separated from the background image, measurements of their number, speed, and so on can be obtained. Typically, background images are extracted from a video stream through image processing because it may be hard to find a frame without any vehicles for normal traffic streams on urban streets. This paper introduces a new method that can quickly extract the background image from traffic video streams for both freeways and intersections in a variety of prevailing traffic conditions. This method has been tested with field data, and the results are promising.

This study applied analytical statistical approaches to investigate some characteristics of traffic accidents. The regression analysis of the injury severities of accidents, and the discrete choice model of different time periods of day when accidents likely occur have been developed. The Poisson regression and Logit model were used by considering the occurrence mechanism of accidents on freeways. Using the observed accident data in Washington State, the accident injury severity model was successfully estimated using a Passion regression. Three variables were found significant in the model. The findings of this study were encouraging. In the studies of Logit model the probability of occurrence of an accident in the different time periods of day including day, dawn, evening and night was expressed by the four utility functions. Both traffic flow and freeway characteristics were included in the model. Compared with most existing models, the new findings were obtained to describe some accidents characteristics.

There is an increasing demand for use of the Global Positioning System (GPS) to navigate or track objects in the forest. However, objects near a GPS receiver antenna, such as tree leaves and branches, can reflect GPS signals and result in large position errors. Canopies in the forest will also block satellite signals and cause the GPS receiver to stop updating data. This is of practical significance for evaluating the performance of GPS in the forest environment. A field test was conducted to understand how large the position errors are and how long the position updates may be deferred under different levels of canopy densities. A digital camera was used to record the canopies over the test site. Image processing techniques, especially Otsu's algorithm, were used and the canopy density was classified into three levels. The ANOVA was used to analyze the effect of canopy density on the GPS position errors. The result shows that the GPS position errors are significantly different under different canopy density levels. The GPS data-update frequency was also analyzed, and the result indicates that the scheduled position update intervals are lengthened due to the existence of forest canopies.