Search

One of the key parameters in radio link planning is the propagation path loss. Most of the existing methods for its prediction are not characterized by a good balance between accuracy, generality, and low computational complexity. To address this problem, a machine learning approach for path loss prediction is presented in this study. The novelty is the proposal of a compound model, which consists of two regression models and one classifier. The first regression model is adequate when a line-of-sight scenario is fulfilled in radio wave propagation, whereas the second one is appropriate for non-line-of-sight conditions. The classification model is intended to provide a probabilistic output, through which the outputs of the regression models are combined. The number of used input parameters is only five. They are related to the distance, the antenna heights, and the statistics of the terrain profile and line-of-sight obstacles. The proposed approach allows creation of a generalized model that is valid for various types of areas and terrains, different antenna heights, and line-of-sight and non line-of-sight propagation conditions. An experimental dataset is provided by measurements for a variety of relief types (flat, hilly, mountain, and foothill) and for rural, urban, and suburban areas. The experimental results show an excellent performances in terms of a root mean square error of a prediction as low as 7.3 dB and a coefficient of determination as high as 0.702. Although the study covers only one operating frequency of 433 M Hz , the proposed model can be trained and applied for any frequency in the decimeter wavelength range. The main reason for the choice of such an operating frequency is because it falls within the range in which many wireless systems of different types are operating. These include Internet of Things (IoT), machine-to-machine (M2M) mesh radio networks, power efficient communication over long distances such as Low-Power Wide-Area Network (LPWAN)—LoRa, etc. [ABSTRACT FROM AUTHOR]

This paper provides a thorough comparative review of deception detection techniques employed for a sample of 400 convicted offenders. It focuses on the utilization of polygraph sensor data as input variables for predicting deception, which are assessed against manual scoring by experts. Three advanced machine learning models, namely Random Forest Regression (RFR), Support Vector Machine (SVM) Regression, and Neural Network Regression (NNR), were employed with the purpose of analysing their predictive efficacy in identifying deception based on physiological responses captured by polygraph sensors. The obtained results indicate that all three algorithms exhibited varying degrees of effectiveness in predicting deceptive behavior. The Random Forest Regression algorithm achieved a Mean Squared Error (MSE) of 0.893 and a coefficient of determination (R²) of 0.091, which highlights its ability to discern key physiological indicators related to deceptive behavior. The Support Vector Machine Regression algorithm showed a competitive performance with a MSE of 0.98 and a R² value of 0.159, which underscores its capability to model non-linear relationships in the context of high-dimensional data. However, the Neural Network Regression algorithm proved to be the best model, with a MSE of 0.894 and a significantly higher R² value of 0.113. This model’s capacity to capture the complex relationships in the context of physiological data allowed it to surpass both RFR and SVM, which indicates its potential for a precise and reliable deception detection. This study provides valuable insights into the advancement of forensic applications with regard to deception detection technologies. Its findings suggest that the Neural Network Regression algorithm, due to its ability to learn complex patterns and relationships related to physiological data, stands out as an optimal choice for accurately identifying deceptive behavior. [ABSTRACT FROM AUTHOR]

Abstract In the railway system, fasteners have the functions of damping, maintaining the track distance, and adjusting the track level. Therefore, routine maintenance and inspection of fasteners are important to ensure the safe operation of track lines. Currently, assessment methods for fastener tightness include manual observation, acoustic wave detection, and image detection. There are limitations such as low accuracy and efficiency, easy interference and misjudgment, and a lack of accurate, stable, and fast detection methods. Aiming at the small deformation characteristics and large elastic change of fasteners from full loosening to full tightening, this study proposes high-precision surface-structured light technology for fastener detection and fastener deformation feature extraction based on the center-line projection distance and a fastener tightness regression method based on neural networks. First, the method uses a 3D camera to obtain a fastener point cloud and then segments the elastic rod area based on the iterative closest point algorithm registration. Principal component analysis is used to calculate the normal vector of the segmented elastic rod surface and extract the point on the centerline of the elastic rod. The point is projected onto the upper surface of the bolt to calculate the projection distance. Subsequently, the mapping relationship between the projection distance sequence and fastener tightness is established, and the influence of each parameter on the fastener tightness prediction is analyzed. Finally, by setting up a fastener detection scene in the track experimental base, collecting data, and completing the algorithm verification, the results showed that the deviation between the fastener tightness regression value obtained after the algorithm processing and the actual measured value RMSE was 0.2196 mm, which significantly improved the effect compared with other tightness detection methods, and realized an effective fastener tightness regression.

In the railway system, fasteners have the functions of damping, maintaining the track distance, and adjusting the track level. Therefore, routine maintenance and inspection of fasteners are important to ensure the safe operation of track lines. Currently, assessment methods for fastener tightness include manual observation, acoustic wave detection, and image detection. There are limitations such as low accuracy and efficiency, easy interference and misjudgment, and a lack of accurate, stable, and fast detection methods. Aiming at the small deformation characteristics and large elastic change of fasteners from full loosening to full tightening, this study proposes high-precision surface-structured light technology for fastener detection and fastener deformation feature extraction based on the center-line projection distance and a fastener tightness regression method based on neural networks. First, the method uses a 3D camera to obtain a fastener point cloud and then segments the elastic rod area based on the iterative closest point algorithm registration. Principal component analysis is used to calculate the normal vector of the segmented elastic rod surface and extract the point on the centerline of the elastic rod. The point is projected onto the upper surface of the bolt to calculate the projection distance. Subsequently, the mapping relationship between the projection distance sequence and fastener tightness is established, and the influence of each parameter on the fastener tightness prediction is analyzed. Finally, by setting up a fastener detection scene in the track experimental base, collecting data, and completing the algorithm verification, the results showed that the deviation between the fastener tightness regression value obtained after the algorithm processing and the actual measured value RMSE was 0.2196 mm, which significantly improved the effect compared with other tightness detection methods, and realized an effective fastener tightness regression. [ABSTRACT FROM AUTHOR]

Abstract A key challenge in quantum photonics today is the efficient and on-demand generation of high-quality single photons and entangled photon pairs. In this regard, one of the most promising types of emitters are semiconductor quantum dots, fluorescent nanostructures also described as artificial atoms. The main technological challenge in upscaling to an industrial level is the typically random spatial and spectral distribution in their growth. Furthermore, depending on the intended application, different requirements are imposed on a quantum dot, which are reflected in its spectral properties. Given that an in-depth suitability analysis is lengthy and costly, it is common practice to pre-select promising candidate quantum dots using their emission spectrum. Currently, this is done by hand. Therefore, to automate and expedite this process, in this paper, we propose a data-driven machine-learning-based method of evaluating the applicability of a semiconductor quantum dot as single photon source. For this, first, a minimally redundant, but maximally relevant feature representation for quantum dot emission spectra is derived by combining conventional spectral analysis with an autoencoding convolutional neural network. The obtained feature vector is subsequently used as input to a neural network regression model, which is specifically designed to not only return a rating score, gauging the technical suitability of a quantum dot, but also a measure of confidence for its evaluation. For training and testing, a large dataset of self-assembled InAs/GaAs semiconductor quantum dot emission spectra is used, partially labelled by a team of experts in the field. Overall, highly convincing results are achieved, as quantum dots are reliably evaluated correctly. Note, that the presented methodology can account for different spectral requirements and is applicable regardless of the underlying photonic structure, fabrication method and material composition. We therefore consider it the first step towards a fully integrated evaluation framework for quantum dots, proving the use of machine learning beneficial in the advancement of future quantum technologies.

This study investigates the application of regression neural networks, particularly the fitrnet model, in predicting the hardness of steels. The experiments involve extensive tuning of hyperparameters using Bayesian optimization and employ 5-fold and 10-fold cross-validation schemes. The trained models are rigorously evaluated, and their performances are compared using various metrics, such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). The results provide valuable insights into the models' effectiveness and their ability to generalize to unseen data. In particular, Model 4208 (8-85-141-1) emerges as the top performer with an impressive RMSE of 1.0790 and an R2 of 0.9900. The model, which was trained with different datasets for nearly 40 steel grades, enables the prediction of hardenability curves, but is limited to the range of the training dataset. The research paper contains an illustrative example that demonstrates the practical application of the developed model in determining the hardenability band for a specific steel grade and shows the effectiveness of the model in predicting and optimizing heat treatment results. [ABSTRACT FROM AUTHOR]

One of the key parameters in radio link planning is the propagation path loss. Most of the existing methods for its prediction are not characterized by a good balance between accuracy, generality, and low computational complexity. To address this problem, a machine learning approach for path loss prediction is presented in this study. The novelty is the proposal of a compound model, which consists of two regression models and one classifier. The first regression model is adequate when a line-of-sight scenario is fulfilled in radio wave propagation, whereas the second one is appropriate for non-line-of-sight conditions. The classification model is intended to provide a probabilistic output, through which the outputs of the regression models are combined. The number of used input parameters is only five. They are related to the distance, the antenna heights, and the statistics of the terrain profile and line-of-sight obstacles. The proposed approach allows creation of a generalized model that is valid for various types of areas and terrains, different antenna heights, and line-of-sight and non line-of-sight propagation conditions. An experimental dataset is provided by measurements for a variety of relief types (flat, hilly, mountain, and foothill) and for rural, urban, and suburban areas. The experimental results show an excellent performances in terms of a root mean square error of a prediction as low as 7.3 dB and a coefficient of determination as high as 0.702. Although the study covers only one operating frequency of 433 MHz, the proposed model can be trained and applied for any frequency in the decimeter wavelength range. The main reason for the choice of such an operating frequency is because it falls within the range in which many wireless systems of different types are operating. These include Internet of Things (IoT), machine-to-machine (M2M) mesh radio networks, power efficient communication over long distances such as Low-Power Wide-Area Network (LPWAN)—LoRa, etc.

Wastewater (WW) analysis is a critical step in various operations, such as the control of a WW treatment facility, and speeding up the analysis of WW quality can significantly improve such operations. This work demonstrates the capability of neural network (NN) regression models to estimate WW characteristic properties such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), ammonia (NH3-N), total dissolved substances (TDS), total alkalinity (TA), and total hardness (TH) by training on WW spectral reflectance in the visible to near-infrared spectrum (400–2000 nm). The dataset contains samples of spectral reflectance intensity, which were the inputs, and the WW parameter levels (BOD, COD, NH3-N, TDS, TA, and TH), which were the outputs. Various NN model configurations were evaluated in terms of regression model fitness. The mean-absolute-error (MAE) was used as the metric for training and testing the NN models, and the coefficient of determination (R2) between the model predictions and true values was also computed to measure how well the NN models predict the true values. The highest R2 (0.994 for training set and 0.973 for testing set) and lowest MAE (0.573 mg/L BOD, 6.258 mg/L COD, 0.369 mg/L NH3-N, 6.98 mg/L TDS, 2.586 m/L TA, and 0.014 mmol/L TH) were achieved when NN models were configured for single-variable output compared to multiple-variables output. Hyperparameter grid-search and k-fold cross-validation improved the NN model prediction performance. With online spectral measurements, the trained neural network model can provide non-contact and real-time estimation of WW quality at minimum estimation error.

Wastewater (WW) analysis is a critical step in various operations, such as the control of a WW treatment facility, and speeding up the analysis of WW quality can significantly improve such operations. This work demonstrates the capability of neural network (NN) regression models to estimate WW characteristic properties such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), ammonia (NH3-N), total dissolved substances (TDS), total alkalinity (TA), and total hardness (TH) by training on WW spectral reflectance in the visible to near-infrared spectrum (400–2000 nm). The dataset contains samples of spectral reflectance intensity, which were the inputs, and the WW parameter levels (BOD, COD, NH3-N, TDS, TA, and TH), which were the outputs. Various NN model configurations were evaluated in terms of regression model fitness. The mean-absolute-error (MAE) was used as the metric for training and testing the NN models, and the coefficient of determination (R2) between the model predictions and true values was also computed to measure how well the NN models predict the true values. The highest R2 (0.994 for training set and 0.973 for testing set) and lowest MAE (0.573 mg/L BOD, 6.258 mg/L COD, 0.369 mg/L NH3-N, 6.98 mg/L TDS, 2.586 m/L TA, and 0.014 mmol/L TH) were achieved when NN models were configured for single-variable output compared to multiple-variables output. Hyperparameter grid-search and k-fold cross-validation improved the NN model prediction performance. With online spectral measurements, the trained neural network model can provide non-contact and real-time estimation of WW quality at minimum estimation error. [ABSTRACT FROM AUTHOR]

Modern cities are full of complex and substantial engineering structures that differ by their geometry, sizes, operating conditions, and technologies used in their construction. During the engineering structures' life cycle, they experience the effects of construction, environmental, and functional loads. Among those structures are bridges and road overpasses. The primary reason for these structures' displacements is land subsidence. The paper addresses a particular case of geospatial monitoring of a road overpass that is affected by external loads invoked by the construction of a new subway line. The study examines the methods of machine learning data analysis and prediction for geospatial monitoring data. The monitoring data were gathered in automatic mode using a robotic total station with a frequency of 30 min, and were averaged daily. Regression analysis and neural network regression with machine learning have been tested on geospatial monitoring data. Apart from the determined spatial displacements, additional parameters were used. These parameters were the position of the tunnel boring machines, precipitation level, temperature variation, and subsidence coefficient. The primary output of the study is a set of prediction models for displacements of the overpass, and the developed recommendations for correctly choosing the prediction model and a set of parameters and hyperparameters. The suggested models and recommendations should be considered an indispensable part of geotechnical monitoring for modern cities. [ABSTRACT FROM AUTHOR]

Loss of geometric accuracy due to shrinkage is a challenge in material extrusion of biological composites using water-based inks, such as the cellulose-chitin biopolymers used here. The shape of 3D printed objects often departs from the intended design geometry due to evaporative loss of water during curing. Moreover, such materials' viscoelastic characteristics result in complex volumetric changes that are difficult to predict and compensate for. We developed a prediction-correction scheme by 3D printing and scanning cylindrical and conic surfaces, computing the geometric deviations between designed and cured artefacts, and training a neural network such that given the machine path for a 3D print, the model can predict shrinkage deformations and apply adjustments on the generating machine paths to proactively compensate it. In this article, we present the shrinkage characteristics of the material used and the results of applying the predictor-correction scheme. The approach substantially improves geometric accuracy, enabling nearly seamless assembly of separately 3D printed parts. Addressing such a fundamental problem of quality control as geometric accuracy may enable the broader adoption of biopolymers and potentially displace the generalised use of synthetic plastics.

This study investigates the application of regression neural networks, particularly the fitrnet model, in predicting the hardness of steels. The experiments involve extensive tuning of hyperparameters using Bayesian optimization and employ 5-fold and 10-fold cross-validation schemes. The trained models are rigorously evaluated, and their performances are compared using various metrics, such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). The results provide valuable insights into the models’ effectiveness and their ability to generalize to unseen data. In particular, Model 4208 (8-85-141-1) emerges as the top performer with an impressive RMSE of 1.0790 and an R2 of 0.9900. The model, which was trained with different datasets for nearly 40 steel grades, enables the prediction of hardenability curves, but is limited to the range of the training dataset. The research paper contains an illustrative example that demonstrates the practical application of the developed model in determining the hardenability band for a specific steel grade and shows the effectiveness of the model in predicting and optimizing heat treatment results.

Loss of geometric accuracy due to shrinkage is a challenge in material extrusion of biological composites using water-based inks, such as the cellulose-chitin biopolymers used here. The shape of 3D printed objects often departs from the intended design geometry due to evaporative loss of water during curing. Moreover, such materials' viscoelastic characteristics result in complex volumetric changes that are difficult to predict and compensate for. We developed a prediction-correction scheme by 3D printing and scanning cylindrical and conic surfaces, computing the geometric deviations between designed and cured artefacts, and training a neural network such that given the machine path for a 3D print, the model can predict shrinkage deformations and apply adjustments on the generating machine paths to proactively compensate it. In this article, we present the shrinkage characteristics of the material used and the results of applying the predictor-correction scheme. The approach substantially improves geometric accuracy, enabling nearly seamless assembly of separately 3D printed parts. Addressing such a fundamental problem of quality control as geometric accuracy may enable the broader adoption of biopolymers and potentially displace the generalised use of synthetic plastics. [ABSTRACT FROM AUTHOR]

As one of the effective measures to reduce energy consumption of high-speed railway, train operation control has gained much attention because of its cost-effectiveness. Conventional approaches to eco-driving generally rely on single-mass train models and lack a thorough investigation into the vehicle cruising process, which can overlook real-world operation conditions and hinder further improvements in energy efficiency. To address these issues, this paper proposes a comprehensive framework for optimizing the cruising speed of high-speed trains (HSTs) based on track conditions to reduce energy consumed by motion resistances (ECMR). Firstly, a multibody dynamics simulation model of the HST is developed and validated to accurately evaluate ECMR under a range of operation scenarios. Simulation results show that the impact of speed on ECMR varies considerably according to track characteristics. Subsequently, a neural network is constructed to derive the regressive relationship between operation conditions and ECMR. On this basis, an optimization model for the cruising speed under different track conditions is formulated to minimize the ECMR of a railway segment with a given operation time. Finally, the effectiveness of the model is validated through case studies, which demonstrate that ECMR can be reduced by up to 27.85% compared to traditional control strategies. The proposed optimization method can be integrated with current train speed profiles to further enhance energy efficiency. This framework has broad applicability for addressing engineering problems related to train operation optimization. [Display omitted] • Energy consumed by motion resistances (ECMR) of high-speed trains (HSTs) is studied. • A dynamic model of the HST is developed to characterize ECMR under various scenarios. • A strong correlation between speed and track properties on ECMR is observed. • A unified framework is built to optimize the cruising speed for energy conservation. • Up to 27.85% of ECMR can be reduced through the cruising speed optimization. [ABSTRACT FROM AUTHOR]

Cheese whey is a difficult and costly wastewater to treat due to its high organic matter and mineral content. Although many management strategies are conducted for whey removal, its use in composting is limited. In this study, the effect of cheese whey in the composting of sewage sludge and poultry waste on compost quality and process efficiency was investigated. Also, valid and consistent simulations were developed with Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Neural Network Regression (NNR) Machine Learning (ML) algorithms. The results of all physicochemical parameters determined that 3% of cheese whey addition for both feedstocks improved the composting process's efficiency and the final product's quality. The best results obtained through hyperparameter tuning showed that Gaussian Process Regression (GPR) was the most effective modeling tool providing realistic simulations. The reliability of these simulations was verified by running the GPR process 50 times. MdAPE demonstrated the validity and consistency of the created process simulations. Moreover, a genetic algorithm was used to optimize these dependent simulations and achieved almost 100% desirability. Optimization studies showed that the effective cheese whey ratios were 3.2724% and 3.1543% for sewage sludge and poultry waste, respectively. Optimization results were compatible with the results of experimental studies. This study provides a new strategy for the recovery of cheese whey as well as a new perspective on the effect of cheese whey on both physicochemical parameters and composting phases and the modeling and optimization processes of the results. [Display omitted] • Cheese whey reduced total nitrogen and organic carbon losses in compost. • The most realistic simulations were achieved by Gaussian Process Regression. • GPR produced estimates with errors 1–2% for T, EC and C/N and <1% for pH, MC and TN. • The effective cheese whey ratio was found 3.27% for SS and 3.15% for PW from GA. • The optimization results by GA were compatible with the experimental results. [ABSTRACT FROM AUTHOR]

Modern cities are full of complex and substantial engineering structures that differ by their geometry, sizes, operating conditions, and technologies used in their construction. During the engineering structures’ life cycle, they experience the effects of construction, environmental, and functional loads. Among those structures are bridges and road overpasses. The primary reason for these structures’ displacements is land subsidence. The paper addresses a particular case of geospatial monitoring of a road overpass that is affected by external loads invoked by the construction of a new subway line. The study examines the methods of machine learning data analysis and prediction for geospatial monitoring data. The monitoring data were gathered in automatic mode using a robotic total station with a frequency of 30 min, and were averaged daily. Regression analysis and neural network regression with machine learning have been tested on geospatial monitoring data. Apart from the determined spatial displacements, additional parameters were used. These parameters were the position of the tunnel boring machines, precipitation level, temperature variation, and subsidence coefficient. The primary output of the study is a set of prediction models for displacements of the overpass, and the developed recommendations for correctly choosing the prediction model and a set of parameters and hyperparameters. The suggested models and recommendations should be considered an indispensable part of geotechnical monitoring for modern cities.

Efficient fast-charging technology is necessary for the extension of the driving range of electric vehicles. However, lithium-ion cells generate immense heat at high-current charging rates. In order to address this problem, an efficient fast charging–cooling scheduling method is urgently needed. In this study, a liquid cooling-based thermal management system equipped with mini-channels was designed for the fast-charging process of a lithium-ion battery module. A neural network-based regression model was proposed based on 81 sets of experimental data, which consisted of three sub-models and considered three outputs: maximum temperature, temperature standard deviation, and energy consumption. Each sub-model had a desirable testing accuracy (99.353%, 97.332%, and 98.381%) after training. The regression model was employed to predict all three outputs among a full dataset, which combined different charging current rates (0.5C, 1C, 1.5C, 2C, and 2.5C (1C = 5 A)) at three different charging stages, and a range of coolant rates (0.0006, 0.0012, and 0.0018 kg·s−1). An optimal charging–cooling schedule was selected from the predicted dataset and was validated by the experiments. The results indicated that the battery module’s state of charge value increased by 0.5 after 15 min, with an energy consumption lower than 0.02 J. The maximum temperature and temperature standard deviation could be controlled within 33.35 and 0.8 °C, respectively. The approach described herein can be used by the electric vehicles industry in real fast-charging conditions. Moreover, optimal fast charging–cooling schedule can be predicted based on the experimental data obtained, that in turn, can significantly improve the efficiency of the charging process design as well as control energy consumption during cooling.

Rainfall plays a main role in managing the water level in the reservoir. The unpredictable amount of rainfall due to the climate change can cause either overflow or dry in the reservoir. In this study, several models and methods were applied to predict the rainfall data in Tasik Kenyir, Terengganu. The comparative study was conducted focusing on developing and comparing several Machine Learning (ML) models, evaluating different scenarios and time horizon, and forecasting rainfall using two types of methods. Data involved for this research consist of taking the average rainfall from 10 stations around the study area using Thiessen polygon to weight the station area and projected rainfall. The forecasting model uses four different ML algorithms, which are Bayesian Linear Regression (BLR), Boosted Decision Tree Regression (BDTR), Decision Forest Regression (DFR) and Neural Network Regression (NNR). On the other hand, the rainfall was predicted on different time horizon by using different ML’s algorithms which is method 1 (M1): Forecasting Rainfall Using Autocorrelation Function (ACF) and method 2 (M2): Forecasting Rainfall Using Projected Error. In M1, the best regression developed for ACF is BDTR since it has the highest coefficient of determination, R2, after tuning the hyperparameter. The results show coefficient between 0.5 and 0.9 with the highest of each scenarios for daily (0.9739693), weekly (0.989461), 10-days (0.9894429) and monthly (0.9998085). In M2, overall model performances show that normalization using LogNormal is preferably giving a good result of each categories except for 10-days with BDTR and DFR are the most acceptable result than NNR and BLR. It is concluded that, two different methods have been applied with different scenarios and different time horizons, and M1 shows a rather high accuracy than M2 using BDTR modeling.

Mapping land cover in highly heterogeneous landscapes is challenging, and classifications have inherent limitations where the spatial resolution of remotely sensed data exceeds the size of small objects. For example, classifications based on 30-m Landsat data do not capture urban or other heterogeneous environments well. This limitation may be overcome by quantifying the subpixel fractions of different land cover types. However, the selection process and transferability of models designed for subpixel land cover mapping across biomes is yet challenging. We asked to what extent (a) locally trained models can be used for sub-pixel land cover fraction estimates in other biomes, and (b) training data from different regions can be combined into spatially generalized models to quantify fractions across global biomes. We applied machine learning regression-based fraction mapping to quantify land cover fractions of 18 regions in five biomes using Landsat data from 2022. We used spectral-temporal metrics to incorporate intra-annual temporal information and compared the performance of local, spatially transferred, and spatially generalized models. Local models performed best when applied to their respective sites (average mean absolute error, MAE, 9–18%), and also well when transferred to other sites within the same biome, but not consistently so for out-of-biome sites. However, spatially generalized models that combined input data from many sites worked very well when analyzing sites in many different biomes, and their MAE values were only slightly higher than those of the respective local models. A weighted training data selection approach, preferring training data with a lower spectral distance to the image data to be predicted, further enhanced the performance of generalized models. Our results suggest that spatially generalized regression-based fraction models can support multi-class sub-pixel fraction estimates based on medium-resolution satellite images globally. Such products would have great value for environmental monitoring in heterogeneous environments and where land cover varies along spatial or temporal gradients. • Spectral-temporal metrics from multi-spectral data distinguish land cover fractions. • Local models accurately map land cover fractions in all biomes. • Spatially generalized models yielded accurate results across sites and biomes. • Weighted training data especially useful for global subpixel fraction maps. [ABSTRACT FROM AUTHOR]

Abstract Alignment-free methods, more time and memory efficient than alignment-based methods, have been widely used for comparing genome sequences or raw sequencing samples without assembly. However, in this study, we show that alignment-free dissimilarity calculated based on sequencing samples can be overestimated compared with the dissimilarity calculated based on their genomes, and this bias can significantly decrease the performance of the alignment-free analysis. Here, we introduce a new alignment-free tool, Alignment-Free methods Adjusted by Neural Network (Afann) that successfully adjusts this bias and achieves excellent performance on various independent datasets. Afann is freely available at https://github.com/GeniusTang/Afann.

Rainfall plays a main role in managing the water level in the reservoir. The unpredictable amount of rainfall due to the climate change can cause either overflow or dry in the reservoir. In this study, several models and methods were applied to predict the rainfall data in Tasik Kenyir, Terengganu. The comparative study was conducted focusing on developing and comparing several Machine Learning (ML) models, evaluating different scenarios and time horizon, and forecasting rainfall using two types of methods. Data involved for this research consist of taking the average rainfall from 10 stations around the study area using Thiessen polygon to weight the station area and projected rainfall. The forecasting model uses four different ML algorithms, which are Bayesian Linear Regression (BLR), Boosted Decision Tree Regression (BDTR), Decision Forest Regression (DFR) and Neural Network Regression (NNR). On the other hand, the rainfall was predicted on different time horizon by using different ML's algorithms which is method 1 (M1): Forecasting Rainfall Using Autocorrelation Function (ACF) and method 2 (M2): Forecasting Rainfall Using Projected Error. In M1, the best regression developed for ACF is BDTR since it has the highest coefficient of determination, R2, after tuning the hyperparameter. The results show coefficient between 0.5 and 0.9 with the highest of each scenarios for daily (0.9739693), weekly (0.989461), 10-days (0.9894429) and monthly (0.9998085). In M2, overall model performances show that normalization using LogNormal is preferably giving a good result of each categories except for 10-days with BDTR and DFR are the most acceptable result than NNR and BLR. It is concluded that, two different methods have been applied with different scenarios and different time horizons, and M1 shows a rather high accuracy than M2 using BDTR modeling. [ABSTRACT FROM AUTHOR]

The performance and reliability of aircraft engine are seriously affected by multiple failures induced by multi-physical loads. Multi-failure probabilistic design is an effective measure to estimate the multi-failure response traits and quantify the multi-failure risk for the improvement of component reliability. In this paper, we propose a neural network regression-distributed collaborative strategy (NNR-DCS) based on a developed two-step error control technique, to improve the efficiency and accuracy of multi-failure probabilistic analysis. We firstly mathematically model NNR-DCS and then introduce the corresponding multi-failure probabilistic framework. With respect to various failure modes such as deformation failure, stress failure and strain failure, the multi-failure probabilistic analysis of a turbine bladed disk is conducted to evaluate the proposed method. From this simulation, we gain the probabilistic distribution features, reliability degree and sensitivity degree of each failure mode and overall failure modes on turbine bladed disk, which provides a useful reference for improving the reliability and performance of aircraft engine. The comparison of methods (Monte Carlo method, RSM, DCRSM, DCFRM, NNR and NNR-DCS) shows that the proposed NNR-DCS holds high efficiency and accuracy for multi-failure probabilistic analysis. The efforts of this study offer an effective way for multi-failure evaluation from a probabilistic perspective and shed light on the multi-objective reliability-based design optimization of complex structures besides turbine bladed disk. [ABSTRACT FROM AUTHOR]

We conducted a statistical analysis of the data, which are included in the regulatory and adaptive capabilities (RAC) estimation algorithm, to obtain the forecasting of maximum boundary (Max.B.) of synchronization range of cardiac and respiratory rhythms, in order to minimize the definition time of RAC of the human body by the method of the cardiorespiratory synchronism (CRS). The following factors have been examined as the basis of the Max.B. forecast: the minimum boundary of the synchronization range, duration of evolution synchronization at the minimum boundary, Initial respiratory rate, Initial heart rate, age, height, weight, gender, menstrual cycle phase (only for women), diastolic and systolic blood pressure values. Linear and quadratic regression models, and neural network techniques - multilayer perceptron and radial basis function network have been analyzed. Neural network regression has shown the best accuracy: the value of synchronization range of cardiac and respiratory rhythms value can be forecast by these factors with the relative error of 20-30%, the probability will be grow up to level of 80-90%.

Abstract: This paper investigates the relation between over-fitting and weight size in neural network regression. The over-fitting of a network to Gaussian noise is discussed. Using re-parametrization, a network function is represented as a bounded function multiplied by a coefficient . This is considered to bound the squared sum of the outputs of at given inputs away from a positive constant , which restricts the weight size of a network and enables the probabilistic upper bound of the degree of over-fitting to be derived. This reveals that the order of the probabilistic upper bound can change depending on . By applying the bound to analyze the over-fitting behavior of one Gaussian unit, it is shown that the probability of obtaining an extremely small value for the width parameter in training is close to one when the sample size is large. [Copyright &y& Elsevier]

The motivation for this paper is to determine the potential economic value of advanced modelling methods for devising trading decision tools for 10-year Government bonds. Two advanced methods are used: time-varying parameter models with the implementation of state space modelling using a Kalman filter and nonparametric nonlinear models with Neural Network Regression (NNR). These are benchmarked against more traditional forecasting techniques to ascertain their potential as a forecasting tool and their economic value as a base for a trading decision tool. The models were developed using data from the UK Gilt market, US T-Bond market and German Bund market. Using in-sample data from April 2001to January 2003to develop the models, their results were assessed using the out-of-sample period of January 2003 to June 2003. Performance evaluation was based upon forecasting accuracy measures and financial criteria using a simulated trading strategy incorporating realistic trading costs. It is concluded that for the time series studied and for the period under investigation, the performance of the advanced models is mixed. While the NNR models have the ability to forecast the 10-year Government bond yield and add economic value as a trading decision tool, the Kalman filter models' performance is not as conclusive. The Kalman filter models outperformed the traditional techniques using forecasting accuracy measures, however they did not perform as well in the simulated trading strategy. [ABSTRACT FROM AUTHOR]