Input Variables Research Articles

Deep Learning based Credit Card Fraudulency Detection System

Abstract: Huge increase in the internet usage has been observed since last decade. It led to the emergence of services like ecommerce, tap and pay systems, online bill payment systems, etc. have proliferated and become more widely used. Due to various online payment options introduced by e- commerce and other numerous websites, the possibility of online fraud has risen drastically. Thus, due to an increase in fraud rates, research on analyzing and detecting fraud in online transactions has begun utilizing various machine learning techniques. The Deep Learning techniques viz., Convolutional Neural Network Architecture is used to detect credit card frauds in the proposed model. The Principal Component Analysis transformation gives a set of numerical input variables as output which are taken as the features to be considered. Due to confidentiality concerns, some of the original characteristics and background information about the data are not entirely disclosed. Features of credit card frauds must be chosen carefully as they play important role when deep learning techniques are used for credit card fraud detection. The TensorFlow and Keras are used for the development of the current proposed model. The proposed model aims to predict credit card frauds with 91.9% accuracy

MINIMIZING TEMPERATURE DEVIATIONS IN RUBBER MIXING PROCESS BY USING ARTIFICIAL NEURAL NETWORKS

ABSTRACT Rubber mixing is a complex manufacturing process that poses challenges for process control due to the high number of control variables, including mixing parameter settings, rheological behavior, compound viscosity, and batch-dependent material variations. Already small deviations from the control variables can influence the compound properties, leading to increased scrap rates. To address these challenges, this paper introduces an artificial intelligence–based approach to enhance process control in rubber mixing by predicting mixing temperatures from input variables. The proposed method uses feedforward neural networks (FFNs) to enable early identification of batch-specific temperature deviations, thereby enabling systematic improvements with each new application. The FFN was trained on a diverse dataset encompassing various rubber recipes and batches. Post-training, the FFN demonstrated remarkable accuracy, achieving a mean absolute percentage error of 1.00% on the training dataset and 1.44% on the validation dataset, thereby showcasing its efficacy in predicting temperature fluctuations within the mixing process. Consequently, the FFN can determine the relevant input variables necessary to achieve specific mixing temperatures, providing a foundation for an automated control system in rubber mixing process. This paper outlines the system architecture of the FFN tailored for rubber mixing and provides a comprehensive overview of the experimental results.

Robust kernel extreme learning machines with weighted mean of vectors and variational mode decomposition for forecasting total dissolved solids

A stable and accurate forecast of water quality parameters is crucial for planning and managing future investment programs. A well-known water quality indicator is the total dissolved solids (TDS), which measures the number of metals, minerals, and salts dissolved in a particular volume of water. This paper introduces an innovative approach, which combines the kernel extreme learning machine (KELM) with the robust weight mean of vectors (INFO) algorithm. The proposed technique stands out for its strategic integration of Boruta-XGBoost (B-XGB) as an exceptional feature selection method, along with variational mode decomposition (VMD) to decompose the input variables and enhance the accuracy of forecasting. The integrated model is termed V-KELM-INFO. INFO is an essential component that incorporates a complex exploration process. It utilizes a weighted mean of vectors, a vector combining operator to enhance population variety, and a localized search operator to expedite convergence. INFO, a potent tool, is included in the V-KELM training stage. It efficiently extracts optimum parameters and dramatically improves the accuracy of monthly TDS predictions at the Idenak station in southwest Iran. Also, the multi-criteria decision-making (MCDM) approach of weighted aggregated sum product assessment (WASPAS) is used to rank models. The V-KELM-INFO model was chosen based on the WASPAS results and statistical metrics (R = 0.962, RMSE = 57.84, WHD = 7.01, and U95 = 160.74) as the best model to forecast the TDS. In addition, the superiority of the proposed model was assessed against the metaheuristic-based V-KELM models, comprising the V-KELM grey wolf optimizer (V-KELM-GWO), V-KELM slime mold algorithm (V-KELM-SMA), and V-KELM equilibrium optimizer (V-KELM-EO). This work provided insight into improving the accuracy of modeling-based methodologies and spurred water quality modeling technology to develop sustainable and clean practices.

Mechanism Analysis and Optimization Design of Exoskeleton Robot with Non-Circular Gear–Pentabar Mechanism

To address the complex structure of existing rod mechanism exoskeleton robots and the difficulty in solving the motion trajectory of multi−rod mechanisms, an exoskeleton knee robot with a differential non−circular gear–pentarod mechanism is designed based on non−circular gears with arbitrary transmission ratios to constrain the degrees of freedom of the R-para-rod mechanism. In this study, the kinematic model of a non-circular gear–five−rod mechanism is established based on motion mapping theory by obtaining the normal motion positions of the human lower limb. An optimization design software for the non-circular gear–five−rod mechanism is developed using the MATLAB 2018b visualization platform, with the non−circular active gear as the sole input variable. A set of ideal parameters is obtained through parameter adjustment and optimal parameter selection, and the corresponding trajectories are compared with human trajectories. The three−dimensional model of the mechanism is established according to the obtained parameters, and the motion simulation of the non−circular gear–five−bar mechanism demonstrates that the mechanism can better reproduce the expected human knee joint motion posture, meeting the working requirements of an exoskeleton knee robot.

Traffic flow prediction: A 3D adaptive multi‐module joint modeling approach integrating spatial‐temporal patterns to capture global features

AbstractThe challenges in citywide traffic flow are intricate, encompassing various factors like temporal and spatial dependencies, holidays, and weather. Despite the complexity, there are still research gaps in effectively incorporating these spatio‐temporal relations through deep learning. Addressing these gaps is crucial for tackling issues such as traffic congestion, public safety, and efficient traffic management within cities. This paper underscores notable research gaps, including the development of models capable of handling both local and global traffic flow patterns, integrating multi‐modal data sources, and effectively managing spatio‐temporal dependencies. In this paper, we proposed a novel model named 3D spatial–temporal‐based adaptive modeling graph convolutional network (3D(STAMGCN)) that addresses for traffic flow data in better periodicity modeling. In contrast to earlier studies, 3D(STAMGCN) approaches the task of traffic flow prediction as a periodic residual learning problem. This is achieved by capturing the input variation between historical time segments and the anticipated output for future time segments. Forecasting traffic flow, as opposed to a direct approach, is significantly simpler when focusing on learning more stationary deviations. This, in turn, aids in the training of the model. Nevertheless, the networks enable residual generation at each time interval through learned variations between future conditions and their corresponding weekly observations. Consequently, this significantly contributes to achieving more accurate forecasts for multiple steps ahead. We executed extensive experiments on two real‐world datasets and compared the performance of our model to state‐of‐the‐art (SOTA) techniques.

APPLICATION OF MACHINE LEARNING MODELS AND GSA METHOD FOR DESIGNING STUD CONNECTORS

The design of stud connectors is aided by determining the relationship between shear strength and the input variables (number, diameter, height, tensile strength and elastic modulus of the studs, and compressive strength and elastic modulus of the concrete) that influence strength. Since strength is nonlinearly related to the influencing variables, which makes the predictions of the relevant empirical equations unreliable, the use of machine learning (ML) models is preferred. The prediction results of eight machine learning models were evaluated, including linear regression (LR1), ridge regression (RR), lasso regression (LR2), back-propagation artificial neural network (BP ANN), genetic algorithm optimized BP ANN (GA-BP ANN), extreme learning machines (ELM), random forests (RF), and support vector machines (SVM). The results show that the GA-BP ANN model is the most accurate model for prediction with a mean absolute percentage error (MAPE) of 6.17% and an R2 of 0.9599. Based on the GA-BP ANN model and the global sensitivity analysis (GSA) method, a new parameter importance analysis method was developed to compare the magnitude of the effect of different input variables on strength. It was found that stud diameter had the greatest effect on shear strength.

Risk analysis of noise-induced hearing loss of workers in the automobile manufacturing industries based on back-propagation neural network model: a cross-sectional study in Han Chinese population.

This study aims to predict the risk of noise-induced hearing loss (NIHL) through a back-propagation neural network (BPNN) model. It provides an early, simple and accurate prediction method for NIHL. Population based, a cross sectional study. Han, China. This study selected 3266 Han male workers from three automobile manufacturing industries. Information including personal life habits, occupational health test information and occupational exposure history were collected and predictive factors of NIHL were screened from these workers. BPNN and logistic regression models were constructed using these predictors. The input variables of BPNN model were 20, 16 and 21 important factors screened by univariate, stepwise and lasso-logistic regression. When the BPNN model was applied to the test set, it was found to have a sensitivity (TPR) of 83.33%, a specificity (TNR) of 85.92%, an accuracy (ACC) of 85.51%, a positive predictive value (PPV) of 52.85%, a negative predictive value of 96.46% and area under the receiver operating curve (AUC) is: 0.926 (95% CI: 0.891 to 0.961), which demonstrated the better overall properties than univariate-logistic regression modelling (AUC: 0.715) (95% CI: 0.652 to 0.777). The BPNN model has better predictive performance against NIHL than the stepwise-logistic and lasso-logistic regression model in terms of TPR, TNR, ACC, PPV and NPV (p<0.05); the area under the receiver operating characteristics curve of NIHL is also higher than that of the stepwise and lasso-logistic regression model (p<0.05). It was a relatively important factor in NIHL to find cumulative noise exposure, auditory system symptoms, age, listening to music or watching video with headphones, exposure to high temperature and noise exposure time in the trained BPNN model. The BPNN model was a valuable tool in dealing with the occupational risk prediction problem of NIHL. It can be used to predict the risk of an individual NIHL.

Mathematical modelling for predicting performance and emissions of a diesel engine fuelled with palm kernel biodiesel–diesel-diethyl carbonate fuel blends

The search for environmentally friendly and sustainable alternatives to conventional diesel fuels has prompted investigations into blends of biodiesel and oxygenated additives like diethyl carbonate. This study focuses on developing mathematical models to predict the performance and emissions of a single-cylinder diesel engine running on palm kernel biodiesel blended with varying proportions of diethyl carbonate. Response Surface Methodology with a general factorial design was employed to create mathematical models for brake thermal efficiency, brake-specific fuel consumption, hydrocarbon, carbon monoxide, and oxides of nitrogen emissions. The two input variables were engine load (five levels) and diethyl carbonate percentage in the fuel blend (five levels). Second-order models were developed, and model fit was evaluated using analysis of variance (ANOVA) at 95% confidence. The developed RSM models showed excellent fit with high R-squared values close to unity, satisfying the model adequacy criteria. ANOVA revealed the models to be statistically significant predictors of the response variables based on load and diethyl carbonate percentage. The optimum operating condition for best performance and emissions balance was found to be 71% load and 10% diethyl carbonate inclusion. The study emphasises the importance of alternative fuels in mitigating environmental pollution and reducing dependence on fossil fuels.

Optimization of the concentration of ozone generated by DBD using PSO algorithm for water treatment process

The water treatment process with ozone is influenced by various operating parameters and environmental factors that can impact its efficiency. In this study, experiments were conducted using a Venturi pumping frame to investigate the effects of three controllable variables: oxygen flow height, applied voltage level, and water flow rate. The tests aimed to develop a mathematical model that accurately represents the relationship between these input variables and the resulting ozone concentration in the treated water. The experimental data was analyzed using the MODDE 5.0 software, a specialized application for statistical modeling and design of experiments. By fitting the data to appropriate model equations, a mathematical model was obtained that quantifies the influence of each variable and their interactions on the ozone concentration response. To optimize the process performance, a particle swarm optimization (PSO) algorithm was employed to extract the best-fit parameters for the mathematical model. PSO is a computational technique inspired by the social behavior of bird flocks or fish schools, utilizing a population of candidate solutions that evolve iteratively to converge on the global optimum solution. In this case, PSO searched for the model parameter values that minimized the error between predicted and experimentally measured ozone concentrations, rapidly converging to an accurate solution. The optimized mathematical model enables predicting the ozone concentration under any combination of oxygen flow height, voltage, and water flow rate within the experimental range. This predictive capability facilitates identifying the optimum operating conditions to maximize ozone concentration, thereby enhancing the efficiency of the water treatment process. The model serves as a valuable tool for process control, monitoring, and optimization, ensuring consistent treatment quality while minimizing resource consumption and operational costs.

Using Optimized Spectral Indices and Machine Learning Algorithms to Assess Soil Copper Concentration in Mining Areas

Soil copper (Cu) contamination in mining areas poses a serious threat to the surrounding environment and human health. Timely determination of Cu concentrations is crucial for the ecological protection of mining areas. Hyperspectral remote sensing technology, with its non-destructive monitoring advantages, is essential for monitoring soil Cu pollution and achieving sustainable agricultural development. Using the hyperspectral technique for assessing soil Cu concentration, four machine learning models (support vector regression (SVR), random forest (RF), partial least squares regression (PLSR), and artificial neural network (ANN)), combined with three types of input variables (the full-band, sensitive bands, and optimized spectral indices (Opt-TBIs)) were employed. The hyperspectral reflectance of 647 soil samples from an abandoned tailings mine in western Inner Mongolia, China was collected. The sensitive bands were extracted using the successive projections algorithms (SPA), and 12 Opt-TBIs were selected. Results showed that the regions with higher soil Cu concentration extracted by SPA and Opt-TBIs were concentrated in the red edge and near-infrared regions. Compared with the full spectrum and SPA-sensitive bands, models based on Opt-TBIs successfully predicted soil Cu concentrations. The Opt-TBIs-RF model provided higher accuracy in estimating soil Cu among the four models. Using only four Opt-TBIs as input variables, the model maintained a stable performance in estimating Cu concentrations in different mining areas (R2Val = 0.72, RPDVal = 1.90). In conclusion, Opt-TBIs as input variables demonstrate good predictive capabilities for soil Cu concentrations in the study area, providing a basis for the formulation of sustainable strategies for soil reclamation and environmental protection in Inner Mongolia.

A constrained machine learning surrogate model to predict the distribution of water-in-oil emulsions in electrostatic fields

Accurately describing the evolution of water droplet size distribution in crude oil is fundamental for evaluating the water separation efficiency in dehydration systems. Enhancing the separation of an aqueous phase dispersed in a dielectric oil phase, which has a significantly lower dielectric constant than the dispersed phase, can be achieved by increasing the water droplet size through the application of an electrostatic field in the pipeline. Mathematical models, while being accurate, are computationally expensive. Herein, we introduced a constrained machine learning (ML) surrogate model developed based on a population balance model. This model serves as a practical alternative, facilitating fast and accurate predictions. The constrained ML model, utilizing an extreme gradient boosting (XGBoost) algorithm tuned with a genetic algorithm (GA), incorporates the key parameters of the electrostatic dehydration process, including droplet diameter, voltage, crude oil properties, temperature, and residence time as input variables, with the output being the number of water droplets per unit volume. Furthermore, we modified the objective function of the XGBoost algorithm by incorporating two penalty terms to ensure the model’s predictions adhere to physical principles. The constrained model demonstrated accuracy on the test set, with a mean squared error of 0.005 and a coefficient of determination of 0.998. The efficiency of the model was validated through comparison with the experimental data and the results of the population balance mathematical model. The analysis shows that the initial droplet diameter and voltage have the highest influence on the model, which aligns with the observed behaviour in the real-world process.

Preoperative glycaemic control, number of pain locations, structural knee damage, self-reported central sensitisation, satisfactionand personal control are predictive of 1-year postoperative pain, and change in pain from pre- to 1-year posttotal knee arthroplasty.

The aim of this study was to identify preoperative predictors for 1-year posttotal knee arthroplasty (TKA) pain and pre- to post-TKA pain difference in knee osteoarthritis (KOA) patients. From March 2018 to July 2023, this prospective longitudinal cohort study enrolled KOA patients awaiting TKA from four hospitals in Belgium and the Netherlands. Different biopsychosocial predictors were assessed preoperatively by questionnaires and physical examinations (input variables). The Knee injury and Osteoarthritis Outcome Score (KOOS) subscale pain was used to measure pain intensity. The absolute KOOS subscale pain score 1-year post-TKA and the difference score (ΔKOOS = 1-year postoperative - preoperative) were used as primary outcome measures (output variables). Two multivariable linear regression analyses were performed. Two hundred and twenty-threeparticipants were included after multiple imputation. Worse absolute KOOS subscale pain scores 1-year post-TKA and negative or closer to zero ΔKOOS subscale pain scores were predicted by self-reported central sensitisation, lower KOA grade and preoperative satisfaction,and higher glycated haemoglobin, number of pain locationsand personal control (adjusted R2 = 0.25). Additional predictors of negative or closer to zero ΔKOOS subscale pain scores were being self-employed, higher preoperative pain and function (adjusted R2 = 0.37). This study reports different biopsychosocial predictors for both outcomes that have filtered out other potential predictors and providevalue for future studies on developing risk assessment tools for the prediction of chronic TKA pain. The protocol is registered at clinicaltrials.gov (NCT05380648) on 13 May 2022. Level II.

Optimized deep neural network to estimate orientation angles for solar photovoltaics intelligent systems

Using a single hidden layer neural network in estimating orientation angles for solar photovoltaics lacks the complexity required to model nonlinear relationships between input variables and the optimal orientation angles for solar photovoltaics. It struggles to generalize well to new and unseen data. More sophisticated neural network architectures such as deep learning with multi-hidden perceptron (MLP) can solve these issues by changing the architecture by deepening the network. Deepening the network will increase complexity, energy consumption, and time complexity. The study uses a novel approach to outperform traditional MLP models with two, three, four, and five hidden layers. An innovative approach was proposed by enhancing a single hidden layer MLP with a quadratic polynomial function, utilizing two robust methodologies, Least Absolute Residuals (LAR) and Bisquare methods. The results demonstrate that these approaches yield significant improvements in Root Mean Square Error (RMSE) and coefficient of determination (R squared). LAR-based MLP showed superiority over both bisquare-based and conventional MLPs methods in R2 and RMSE, ranging from 1.13 to 1.18 and 2.53 to 3.06, respectively. The study outperformed conventional MLP architectures with five hidden layers regarding accuracy and efficiency. The proposed model offers a more effective and less complex solution for data prediction tasks.

SIMULATION OF TRANSIENT PROCESS ON WWER-1000 USING COUPLED APPROACH

Research and modeling of thermohydraulic processes in NPP equipment is necessary for a deep and improved justification of reliable and safe operation of nuclear installations. Taking into account the accumulated experience and the latest technologies, the study of thermohydraulic processes is developing in the direction of applying innovative approaches to conducting both experimental studies and expanding the calculation capabilities of thermohydraulic codes. The article presents an approach to the decomposition of the thermohydraulic model of the WWER-1000 reactor plant on submodels, which are calculated in coupling with separate instances of the system thermohydraulic code RELAP5/Mod 3.2. Coupling is implemented using a specially developed module. This approach allows you to "bypass" the internal limitations of RELAP5 in the number of components (hydrodynamic volumes, thermal structures, "trips" and control variables) that can be used for computational analysis. As a result, it becomes possible to achieve a higher degree of detail for the entire reactor installation. In addition, the article confirms the possibility of applying the previously developed coupling module to such a non-equilibrium model as a nuclear reactor model. At the same time, the use of the RELAP5 model as a reactor model makes it possible to avoid inaccuracies that arise when applying assumptions about the transverse profiles of the input variables of the CFX model, as well as to perform the validation of the RELAP5 model of the RU loops. Coupled calculation of stationary and transient mode of reactor installation WWER-1000 is performed. The reliability of calculation results obtained in conjugation is evaluated by comparing them with the results of autonomous calculations in RELAP5. The analysis of the results of computational modeling confirms the possibility of applying the developed coupling module to such a non-equilibrium model as a model of a nuclear reactor, and also confirms the correctness of the RELAP5 model of the loops of the reactor plant obtained on the basis of the modification of the full-fledged RELAP5 model.

Remote sensing estimation of δ15NPN in the Zhanjiang Bay using Sentinel-3 OLCI data based on machine learning algorithm

Remote sensing estimation of δ15NPN in the Zhanjiang Bay using Sentinel-3 OLCI data based on machine learning algorithm

Forecasting Hydrogen Vehicle Refuelling for Sustainable Transportation: A Light Gradient-Boosting Machine Model

Forecasting Hydrogen Vehicle Refuelling for Sustainable Transportation: A Light Gradient-Boosting Machine Model

Artificial neural network, machine learning modelling of compressive strength of recycled coarse aggregate based self-compacting concrete.

This research study aims to understand the application of Artificial Neural Networks (ANNs) to forecast the Self-Compacting Recycled Coarse Aggregate Concrete (SCRCAC) compressive strength. From different literature, 602 available data sets from SCRCAC mix designs are collected, and the data are rearranged, reconstructed, trained and tested for the ANN model development. The models were established using seven input variables: the mass of cementitious content, water, natural coarse aggregate content, natural fine aggregate content, recycled coarse aggregate content, chemical admixture and mineral admixture used in the SCRCAC mix designs. Two normalization techniques are used for data normalization to visualize the data distribution. For each normalization technique, three transfer functions are used for modelling. In total, six different types of models were run in MATLAB and used to estimate the 28th day SCRCAC compressive strength. Normalization technique 2 performs better than 1 and TANSING is the best transfer function. The best k-fold cross-validation fold is k = 7. The coefficient of determination for predicted and actual compressive strength is 0.78 for training and 0.86 for testing. The impact of the number of neurons and layers on the model was performed. Inputs from standards are used to forecast the 28th day compressive strength. Apart from ANN, Machine Learning (ML) techniques like random forest, extra trees, extreme boosting and light gradient boosting techniques are adopted to predict the 28th day compressive strength of SCRCAC. Compared to ML, ANN prediction shows better results in terms of sensitive analysis. The study also extended to determine 28th day compressive strength from experimental work and compared it with 28th day compressive strength from ANN best model. Standard and ANN mix designs have similar fresh and hardened properties. The average compressive strength from ANN model and experimental results are 39.067 and 38.36 MPa, respectively with correlation coefficient is 1. It appears that ANN can validly predict the compressive strength of concrete.

Prediction of energy consumption for manufacturing small and medium-sized enterprises (SMEs) considering industry characteristics

There has been a growing demand for energy consumption statistics in the manufacturing industry to establish national energy and greenhouse gas policies. Despite its importance, the Korean government faces significant challenges in collecting energy data at a facility level in a precise and timely manner. To address the lack of timely data, this paper employs machine-learning models to predict the annual total energy consumptions of each manufacturing facility. We first designed four prediction models that take into account the characteristics and energy consumption behaviors of industry sub-sectors. As input variables, these prediction models mainly included electricity consumption, employee size, energy types, gas consumption and other accessible data. Finally, we conducted numerical experiments on approximately 100,000 facilities and evaluated the prediction performance of various machine-learning algorithms such as linear regression, decision tree regression, random forest regression, gradient boost regression, and extreme gradient boosting regression. The numerical experiments provided insights into which model and algorithm offer the best prediction performance for each industry sub-sector. In addition, we identified the important variables for predicting total energy consumption, revealing that not only electricity but also various other energy sources and variables representing industry-specific characteristics play a crucial role in improving prediction performance.

PERSONALITY PREDICTION SYSTEM

Personality prediction is a challenging yet crucial task in various fields such as psychology, human resources, and marketing. In this study, we propose a questionnaire-based approach using the random forest algorithm to predict personality traits. The questionnaire is designed to gather information related to the personality traits: openness, conscientiousness, extraversion, agreeableness, and neuroticism. The dataset used for this study consists of responses from individuals who completed the questionnaire. Responses to specific questions are used as input variables for the random forest algorithm. The algorithm is trained on a portion of the dataset and then tested on the remaining portion to evaluate its performance in predicting personality traits. Our results show that the random forest algorithm achieves high accuracy in predicting personality traits, outperforming other machine learning algorithms such as logistic regression and support vector machines. This approach has the potential to be used in various applications, such as personalized marketing, recommendation systems, and mental health assessment. Key Words: Personality prediction, Random forest algorithm, personality traits, Questionnaire-based approach.

Using Transfer Learning and XGBoost for Early Detection of Fires in Offshore Wind Turbine Units

To improve the power generation efficiency of offshore wind turbines and address the problem of high fire monitoring and warning costs, we propose a data-driven fire warning method based on transfer learning for wind turbines in this paper. This paper processes wind turbine operation data in a SCADA system. It uses an extreme gradient-boosting tree (XGBoost) algorithm to build an offshore wind turbine unit fire warning model with a multiparameter prediction function. This paper selects some parameters from the dataset as input variables for the model, with average cabin temperature, average outdoor temperature, average cabin humidity, and average atmospheric humidity as output variables. This paper analyzes the distribution information of input and output variables and their correlation, analyzes the predicted difference, and then provides an early warning for wind turbine fires. This paper uses this fire warning model to transfer learning to different models of offshore wind turbines in the same wind farm to achieve fire warning. The experimental results show that the prediction performance of the multiparameter is accurate, with an average MAPE of 0.016 and an average RMSE of 0.795. It is better than the average MAPE (0.051) and the average RMSE (2.020) of the prediction performance of a backpropagation (BP) neural network, as well as the average MAPE (0.030) and the average RMSE (1.301) of the prediction performance of random forest. The transfer learning model has good prediction performance, with an average MAPE of 0.022 and an average RMSE of 1.469.