Dimensionality Reduction Process Research Articles

Neuronal Network-Based Multiplicatively Gait Feature Eradication and Detection

The accuracy of gait detection is affected by external influences such as intricate backdrop and coverings. To address this issue, this research presents an artificial neuronal network-based multiplicatively gait feature eradication and detection procedure. The proposed procedure uses a imaging device and a pyroelectric infrared sensor to separately acquire gait data. The skeletal attribute factor, peak variation attribute factor from the imaging device, and frequency spectrum attribute factor from the pyroelectric infrared sensor are extracted and combined for dimensionality reduction and signal processing. Finally, the merged attributes are classified and identified using a BP neuronal network as the classifier. The proposed procedure is evaluated for its detection accuracy.

Building Engineering Cost Prediction Model Based on TSNE and Improved Grey Correlation Algorithm

With the development of digitalization in the construction industry, the complexity of construction projects is increasing. In response to the widely valued cost and quality management of construction projects, improved algorithms based on machine vision learning can effectively cope with the increasingly complex cost prediction of construction projects. In the entire model design, how to complete the selection and processing of key indicators for construction projects is currently a key issue that needs to be urgently solved. This article studied the use of TSNE (T-Distributed Stochastic Neighbor Embedding) and improved grey correlation algorithm to design a cost prediction model, aiming to improve the prediction accuracy of construction project costs and reduce system prediction time. This article analyzed the cost index system and simulation experiments of construction projects, applied machine learning based TSNE and improved grey correlation algorithm, combined index dimensionality reduction processing and construction project unit test set samples, and discussed the following conclusions based on the data results: In the context of intelligent large-scale infrastructure, the construction project cost prediction model using TSNE and improved grey correlation algorithm had higher prediction accuracy than traditional prediction methods in all construction project samples, with an average improvement of 5.1%. At the same time, in terms of model prediction efficiency, the improved algorithm was better than traditional prediction methods, with an average improvement of 12.75%. This indicated that the construction project cost prediction model based on TSNE and improved grey correlation algorithm has a relatively good prediction effect overall.

Research on Stock Price Prediction of BP Neural Network Based on Factor Analysis

In this thesis, the real estate industry is focused on, the financial data of 27 listed real estate companies in China is chosen, and the method of factor analysis is used to carry out the dimensionality reduction processing for the collected 20 variables. Finally, six main factors are extracted. Besides, the stock prices before and after the factor analysis are predicted and studied respectively based on the BP neural network prediction model. In line with the results, the better prediction accuracy for the BP neural network model established with the main factor determined by factor analysis as the independent variable and the stock price as the dependent variable is embodied. In contrast with the research carried out by other scholars, the BP neural network is established from more dimensions, so the prediction accuracy is greatly enhanced.

Class-Constrained t-SNE: Combining Data Features and Class Probabilities.

Data features and class probabilities are two main perspectives when, e.g., evaluating model results and identifying problematic items. Class probabilities represent the likelihood that each instance belongs to a particular class, which can be produced by probabilistic classifiers or even human labeling with uncertainty. Since both perspectives are multi-dimensional data, dimensionality reduction (DR) techniques are commonly used to extract informative characteristics from them. However, existing methods either focus solely on the data feature perspective or rely on class probability estimates to guide the DR process. In contrast to previous work where separate views are linked to conduct the analysis, we propose a novel approach, class-constrained t-SNE, that combines data features and class probabilities in the same DR result. Specifically, we combine them by balancing two corresponding components in a cost function to optimize the positions of data points and iconic representation of classes - class landmarks. Furthermore, an interactive user-adjustable parameter balances these two components so that users can focus on the weighted perspectives of interest and also empowers a smooth visual transition between varying perspectives to preserve the mental map. We illustrate its application potential in model evaluation and visual-interactive labeling. A comparative analysis is performed to evaluate the DR results.

A CNN-LSTM-PSO tool wear prediction method based on multi-channel feature fusion

In order to achieve predictive maintenance of CNC machining tools and to be able to change tools intelligently before tool wear is at a critical threshold, a CNN-LSTM tool wear prediction model based on particle swarm algorithm (PSO) optimization with multi-channel feature fusion is proposed. Firstly, the raw signals of seven channels of the machining process are collected using sensor technology and processed for noise reduction; secondly, the time-domain, frequency-domain and time-frequency-domain features of each channel signal are extracted, and a sample data set of spatio-temporal correlation of traffic flow is constructed by dimensionality reduction processing and information fusion of the above features; finally, the data set is input to the CNN-LSTM-PSO model for training and testing. The results show that the CNN-LSTM-PSO model can effectively predict tool wear with an average absolute error MAE value of 0.5848, a root mean square error RMSE value of 0.7281, and a coefficient of determination R2 value of 0.9964; and compared with the BP model, CNN model, LSTM model and CNN-LSTM model, its tool wear prediction accuracy improved by 7.56%, 2.60%, 2.98%, and 1.63%, respectively.

Implementation of K-Means Clustering Method for Trend Analysis of Thesis Topics (Case Study: Faculty of Computer Science, University of Jember)

The development of information technology causes a large number of digital documents, especially thesis documents, so that it can create opportunities for students to take the same and not varied topics. Thesis documents can be grouped by topic by identifying the abstract section. The results of the grouping can be seen with the trend with data visualization so that it can be analyzed to find out the trend of each topic. Retrieval of data in the repository of the University of Jember through a web scraping process as many as 490 thesis documents for students of the Faculty of Computer Science, University of Jember. The preprocessing stage is carried out by text mining methods which include cleaning, filtering, stemming, and tokenizing. Then calculate the weight of each word with the Term Frequency - Inverse Document Frequency algorithm, followed by the dimension reduction process using the Principal Component Analysis algorithm, which is normalized by Z-Score first. The outliers removal process is carried out before classifying documents. Furthermore, document grouping uses the K-Means Clustering method with Cosine Similarity as the distance calculation and the Silhouette Coefficient algorithm as a test. The test results were carried out with various k values and the optimal value was obtained at k = 2 with a Silhouette value of 0.80. Then the topic detection uses the Latent Dirichlet Allocation algorithm for each cluster that has been formed. Each cluster is visualized with a line chart and Trend Linear algorithm and analyzed to find out the trend. From the results of the analysis, it can be concluded that the topic of Decision Support System Development is trending down, and the topic of IT Performance Measurement and Forecasting is trending up. It can be concluded that the topic of Decision Support System Development needs to be reduced so that other topics can emerge.

An Optimized Double-Nested Anti-Missile Force Deployment Based on the Deep Kuhn–Munkres Algorithm

In view of a complex multi-factor interaction relationship and high uncertainty of a battlefield environment in the anti-missile troop deployment, this paper analyzes the relationships between the defending stronghold, weapon system, incoming target, and ballistic missile. In addition, a double nested optimization architecture is designed by combining deep learning hierarchy concept and hierarchical dimensionality reduction processing. Moreover, a deployment model based on the double nested optimization architecture is constructed with the interception arc length as an optimization goal and based on the basic deployment model, kill zone model, and cover zone model. Further, by combining the target full coverage adjustment criterion and depth-first search, a deep Kuhn–Munkres algorithm is proposed. The model is validated by simulations of typical scenes. The results verify the rationality and feasibility of the proposed model, high adaptability of the proposed algorithm. The research of this paper has important enlightenment and reference function for solving the force deployment optimization problems in uncertain battlefield environment.

The Improvement of Artificial Neural Network Accuracy Using Principle Component Analysis Approach

An important problem in a classification system is how to get good accuracy results. A way to increase the accuracy of a classifier system is to improve the number of input data attributes. Improving the number of input data attributes can be done using the Principal Component Analysis (PCA) method. The aim of this research is to reduce the number of input data attributes to increase the accuracy in a mushroom classification system. The research method used in this study started from collecting datasets from Kaggle.com related to mushroom-classification, then the data visualization process was carried out using pie charts then a dimension reduction process was carried out to reduce the number of variables using the PCA method. The next step is the training and testing of the artificial neural network. The architecture of artificial neural network used is backward error propagation with the number of hidden layers as much as 2 layers with the number of cells as many as 3 and 2. The training data used is 80%, while the testing data is 20%. Based on the test results, obtained an accuracy of 100% with 150,000 iterations and using 11 input variables from 22 existing input variables. By adding Principal Component Analysis part of the development that can improve the accuracy and performance of Artificial Neural Networks

Multi-objective decision support system for large-scale network pavement maintenance and rehabilitation management to enhance sustainability

Process of sustainable pavement maintenance and rehabilitation (M&R) management can be conducted using various technological and organizational systems. Such systems must meet the implementation conditions of a given project involving pavement M&R, and trade-offs between economic, environmental, and social impacts from various aspects involved. Hence, this study examines a systematic framework for large-scale pavement network decision-making considering sustainability dimensions, constructs a multi-stage decision model considering different modes and options using attribute mapping techniques, and develops a decision support system that (1) determines comprehensive sustainability drivers and quantification methods in the field of pavement management to evaluate the sustainability characteristics of predetermined M&R strategies; and (2) allows the execution of decision space dimensionality reduction process for any given network scales and M&R techniques dimensions considering spatial-temporal distribution, then generate a specific set of sustainable budget allocation and M&R strategy combination scenarios using a redesigned multi-objective optimization model, which can coordinate segment-level and network-level management. A case study was conducted to demonstrate the application of the proposed system. Results show that the approach can significantly reduce the computational cost compared to the scenario without dimensionality reduction, and also facilitates the improvement of agency costs, environmental impacts, as well as user costs from the life cycle perspective. Furthermore, the balance of M&R budget allocation depends on the mapped attributes selected in the dimensionality reduction process, the sustainability and reliability of the decision results can be improved by selecting as many attributes as possible while satisfying the pavement management system. The study provides an exciting opportunity to support decision-making for large-scale transportation infrastructure management.

HM-YOLOv5: A fast and accurate network for defect detection of hot-pressed light guide plates

Due to the complex texture background, uneven brightness, various defect sizes and types of hot-pressed light guide plate (LGP) images, the HM-YOLOv5 network for fast and accurate defect detection of hot-pressed LGPs is developed and evaluated in this paper. First, a hybrid attention module (HAM) is constructed by combining the spatial attention of the convolutional block attention module(CBAM) with the channel attention of the efficient channel attention network(ECA-Net), and is introduced behind each C3 module of the backbone network. HAM does not require dimensionality reduction processing, and can better fuse channel and spatial information to focus on defects targets. Therefore, the HAM can enhance the defect feature extraction ability of the network. Second, a multi-expansion convolution module (MCM) is constructed based on the pyramid structure of YOLOv5. MCM can improve the target receptive field, enrich contextual information, reduce loss of information during downsampling and improve the defect detection ability Finally, a self-built dataset is constructed by using the images of hot-pressed LGPs collected from industrial sites, and many experiments are performed using this database. Experimental results show that the mean average precision (mAP) of the network is 98.9%, especially for white point and dark line defects improved by 2.7% and 2.0% respectively, and that the detection speed can reach 417 frames per second(Fps). Compared with the mainstream surface defect detection algorithm, the accuracy and detection time are significantly improved. Moreover, the accuracy and real-time performance meet the industrial detection requirements.

Short-term day-ahead photovoltaic output forecasting using PCA-SFLA-GRNN algorithm

The work of forecasting solar power is becoming more crucial with directives to regulate the quality of the power and increase the system’s reliability as photovoltaic (PV) sites are being integrated into the architecture of power systems at an increasing rate. This study proposes a metaheuristic model for short-term photovoltaic power forecasting that includes shuffled frog leaping algorithm (SFLA), principal component analysis (PCA), and generalized regression neural network (GRNN). In this model, GRNN is implemented to analyze the input parameters after the dimension reduction process, and its parameters get optimized with the help of the SFLA, which has the advantage of fast convergence speed as well as searching ability, whereas PCA techniques are implemented to diminish the dimension of meteorological conditions. This hybrid model achieves day-ahead short-term forecasting, as shown in an experimental case of a Bhadla Solar Park installed in Gujarat, India. The accuracy of the proposed model obtained a mean absolute error (nMAE) of 2.3325, and a root mean square error (RMSE) of 129.425. Similarly, the error in forecasting obtained by the proposed method results in nMAE = 2.977 and RMSE = 160.92. The output results obtained surpassed all other hybrid models used for comparison in this study.

Sentiment analysis on social media tweets using dimensionality reduction and natural language processing

AbstractSocial media has been embraced by different people as a convenient and official medium of communication. People write or share messages and attach images and videos on Twitter, Facebook and other social media platforms. It therefore generates a lot of data that is rich in sentiments. Sentiment analysis has been used to determine the opinions of clients, for instance, relating to a particular product or company. Lexicon and machine learning approaches are the strategies that have been used to analyze these sentiments. The performance of sentiment analysis is, however, distorted by noise, the curse of dimensionality, the data domains and the size of data used for training and testing. This article aims at developing a model for sentiment analysis of social media data in which dimensionality reduction and natural language processing with part of speech tagging are incorporated. The model is tested using Naïve Bayes, support vector machine, and K‐nearest neighbor algorithms, and its performance compared with that of two other sentiment analysis models. Experimental results show that the model improves sentiment analysis performance using machine learning techniques.

Meter Wave Polarization-Sensitive Array Radar for Height Measurement Based on MUSIC Algorithm.

Obtaining good measurement performance with meter wave radar has always been a difficult problem. Especially in low-elevation areas, the multipath effect seriously affects the measurement accuracy of meter wave radar. The generalized multiple signal classification (MUSIC) algorithm is a well-known measurement method that dose not require decorrelation processing. The polarization-sensitive array (PSA) has the advantage of polarization diversity, and the polarization smoothing generalized MUSIC algorithm demonstrates good angle estimation performance in low-elevation areas when based on a PSA. Nevertheless, its computational complexity is still high, and the estimation accuracy and discrimination success probability need to be further improved. In addition, it cannot estimate the polarization parameters. To solve these problems, a polarization synthesis steering vector MUSIC algorithm is proposed in this paper. First, the MUSIC algorithm is used to obtain the spatial spectrum of the meter wave PSA. Second, the received data are properly deformed and classified. The Rayleigh–Ritz method is used to decompose the angle to realize the decoupling of polarization and the direction of the arrival angle. Third, the geometric relationship and prior information of the direct wave and the reflected wave are used to continue dimension reduction processing to reduce the computational complexity of the algorithm. Finally, the geometric relationship is used to obtain the target height measurement results. Extensive simulation results illustrate the accuracy and superiority of the proposed algorithm.

Local homeomorphism regularized non-negative Tucker decomposition of tensor data

Non-negative Tucker decomposition of tensor data has a wide range of applications in machine learning. With the non-negative constraints, tensor data can be decomposed into the mode product of a core tensor and a series of projection matrices. The core tensor usually is be regarded as the low-dimensional representation of the original tensor. The process of dimensionality reduction preserves the global properties of tensor data. But many applications in machine learning expect the continuous dependencies of data local feature to remain unchanged during dimensionality reduction. To this end, this paper proposes a local homeomorphism regularized non-negative Tucker decomposition algorithm for tensor data. The proposed method introduce a local homeomorphism regularized term to tensor non-negative Tucker decomposition constrain for effectively preserving the global and local characters of tensor. Experiments on four commonly used real data sets and six compared algorithms have demonstrate the well performance of the proposed algorithms.

Development of Machine Learning and Medical Enabled Multimodal for Segmentation and Classification of Brain Tumor Using MRI Images

The improper and excessive growth of brain cells may lead to the formation of a brain tumor. Brain tumors are the major cause of death from cancer. As a direct consequence of this, it is becoming more challenging to identify a treatment that is effective for a specific kind of brain tumor. The brain may be imaged in three dimensions using a standard MRI scan. Its primary function is to examine, identify, diagnose, and classify a variety of neurological conditions. Radiation therapy is employed in the treatment of tumors, and MRI segmentation is used to guide treatment. Because of this, we are able to assess whether or not a piece that was spotted by an MRI is a tumor. Using MRI scans, this study proposes a machine learning and medically assisted multimodal approach to segmenting and classifying brain tumors. MRI pictures contain noise. The geometric mean filter is utilized during picture preprocessing to facilitate the removal of noise. Fuzzy c-means algorithms are responsible for segmenting an image into smaller parts. The identification of a region of interest is facilitated by segmentation. The GLCM Grey-level co-occurrence matrix is utilized in order to carry out the process of dimension reduction. The GLCM algorithm is used to extract features from photographs. The photos are then categorized using various machine learning methods, including SVM, RBF, ANN, and AdaBoost. The performance of the SVM RBF algorithm is superior when it comes to the classification and detection of brain tumors.

Advances in Hyperspectral Image Classification with a Bottleneck Attention Mechanism Based on 3D-FCNN Model and Imaging Spectrometer Sensor

Deep learning approaches have significantly enhanced the classification accuracy of hyperspectral images (HSIs). However, the classification process still faces difficulties such as those posed by high data dimensions, large data volumes, and insufficient numbers of labeled samples. To enhance the classification accuracy and reduce the data dimensions and training needed for labeled samples, a 3D fully convolutional neural network (3D-FCNN) model was developed by including a bottleneck attention module. In such a model, the convolutional layer replaces the downsampling layer and the fully connected layer, and 3D full convolution is adopted to replace the commonly used 2D and 1D convolution operations. Thus, the loss of data in the dimensionality reduction process is effectively avoided. The bottleneck attention mechanism is introduced in the FCNN to reduce the redundancy of information and the number of labeled samples. The proposed method was compared to some advanced HSI classification approaches with deep networks, and five common HSI datasets were employed. The experiments showed that our network could achieve considerable classification accuracies by reducing the data dimensionality using a small number of labeled samples, thereby demonstrating its potential merits in the HSI classification process .

Big Data Analysis and Application of Liver Cancer Gene Sequence Based on Second-Generation Sequencing Technology.

In big data analysis with the rapid improvement of computer storage capacity and the rapid development of complex algorithms, the exponential growth of massive data has also made science and technology progress with each passing day. Based on omics data such as mRNA data, microRNA data, or DNA methylation data, this study uses traditional clustering methods such as kmeans, K-nearest neighbors, hierarchical clustering, affinity propagation, and nonnegative matrix decomposition to classify samples into categories, obtained: (1) The assumption that the attributes are independent of each other reduces the classification effect of the algorithm to a certain extent. According to the idea of multilevel grid, there is a one-to-one mapping from high-dimensional space to one-dimensional. The complexity is greatly simplified by encoding the one-dimensional grid of the hierarchical grid. The logic of the algorithm is relatively simple, and it also has a very stable classification efficiency. (2) Convert the two-dimensional representation of the data into the one-dimensional representation of the binary, realize the dimensionality reduction processing of the data, and improve the organization and storage efficiency of the data. The grid coding expresses the spatial position of the data, maintains the original organization method of the data, and does not make the abstract expression of the data object. (3) The data processing of nondiscrete and missing values provides a new opportunity for the identification of protein targets of small molecule therapy and obtains a better classification effect. (4) The comparison of the three models shows that Naive Bayes is the optimal model. Each iteration is composed of alternately expected steps and maximal steps and then identified and quantified by MS.

Drivers’ attitudes to road safety in the South East Asian cities of Jakarta and Hanoi: Socio-economic and demographic characterisation by Multiple Correspondence Analysis

This study investigates drivers’ attitudes towards road safety in Jakarta and Hanoi. A comprehensive analysis of two datasets was achieved by multi-faceted analytical approaches of Multiple Correspondence Analysis (MCA), Cluster analysis (Hierarchical and Two-step) (CA), Principal Component Analysis (PCA) and Multinomial Logistic Regression (MLR). MCA and CA were used to segment the drivers based on their sociodemographic characteristics and driving related information where three clusters were identified from each dataset. There were 60+ attitudinal statements in the datasets, therefore dimension reduction process was conducted using PCA. Five Principal Components were generated from each dataset, four of them common to Jakarta and Hanoi (“Safe driving practices and behaviour”, “Road safety enforcement and education”, “Driver behaviour at signalised junctions”, “Road infrastructure and roadside facilities”), the fifth component being unique to each city (“Road infrastructure design issues” for Hanoi and “Type of motorised vehicles” for Jakarta). Finally, MLR was used to investigate how the level of perception on the identified components varied across sociodemographic clusters in each city. Employed young adults in Jakarta perceive that “Driver behaviour at signalised junctions” directly influences road safety compared to teenage and young adults who are in education. The reverse is true for Hanoi. Employed young drivers in both cities are less likely to recognise the importance of “Road safety enforcement and education” compared to teenage and young drivers in education. The study shows that the perception of road safety among drivers varies based on various factors including their sociodemographic traits and driving experience.

Early Warning of Enterprise Financial Risk Based on Decision Tree Algorithm.

To improve enterprise financial early warning, we propose an algorithm based on a decision tree. According to the shortcomings and defects of the classical algorithm and the traditional decision tree algorithm, in the ordinary decision tree improved algorithm based on PCA, there is a problem that the representativeness of the data after dimensionality reduction processing are not high, resulting in the fact that the accuracy of the algorithm can be improved slightly after multiple data runs. Based on the classical algorithm, attribute eigenvalues before classification are extracted twice, and the amount of data to be classified is calculated. That is, the most important attributes of the original data are selected. After the subtree is established, the dimension reduction and merging selection of the data are performed, and the improved algorithm is verified by using three data sets in the UCI database. The results show that the average accuracy in the three datasets is 94.6%, which is improved by 1.6% and 0.6% for the traditional classical algorithm and the ordinary PCA decision tree optimization algorithm, respectively. PCA-based decision tree algorithms can improve the accuracy of the results to some extent, which is of practical importance. In the future, a classic algorithm improved for secondary modeling will be used to obtain a more efficient decision tree model. The decision tree algorithm has been proven to recognize an early warning of an enterprise's financial risks, which enhances the effectiveness of an enterprise's early financial warning.

Quantitative analysis of typical fluoroquinolone antibiotics in livestock feed using terahertz spectrum

The antibiotic residues are increasing concerns in agricultural and food products, an effective, rapid, and accurate detection method of antibiotics is still in exploration stage for applications. In this study, we employed terahertz (THz) spectroscopy as a tool to find its feasibility to detect quinolone antibiotics in livestock feed (LSF) quantitatively. A total of 220 samples with pefloxacin (PEF) and fleroxacin (FLE) in LSF were prepared separately and the THz spectrum of each sample was measured using a TERA K15 THz-TDS system. Then, the absorption characteristics of PEF and FLE in THz band were analyzed. For the binary and ternary mixtures, successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS) algorithm, and the combination algorithm (CARS-SPA) were selected as dimensionality reduction process methods, and back propagation neural network (BPNN) was selected to establish a model for the quantitative analysis. The results showed, PEF has two typical absorption peaks at 0.775 and 0.988 THz, and FLE has two typical absorption peaks at 0.919 and 1.088 THz, which implies the features can be used for qualitative identification. For the quantitative analysis, The CARS-BPNN model achieved the best predictive ability based on the prediction set, and the correlation coefficients can reach 0.98 for both the binary and ternary mixtures. The results also showed that predictive models of the ternary mixtures have better performance than that of the binary mixtures. The research provides theoretical and technical reference for practical detection of antibiotic residues in livestock and poultry industry.