Training Data Research Articles

Mitigating Recommendation Biases via Group-Alignment and Global-Uniformity in Representation Learning

Collaborative Filtering (CF) plays a crucial role in modern recommender systems, leveraging historical user-item interactions to provide personalized suggestions. However, CF-based methods often encounter biases due to imbalances in training data. This phenomenon makes CF-based methods tend to prioritize recommending popular items and performing unsatisfactorily on inactive users. Existing works address this issue by rebalancing training samples, reranking recommendation results, or making the modeling process robust to the bias. Despite their effectiveness, these approaches can compromise accuracy or be sensitive to weighting strategies, making them challenging to train. Therefore, exploring how to mitigate these biases remains in urgent demand. In this paper, we deeply analyze the causes and effects of the biases and propose a framework to alleviate biases in recommendation from the perspective of representation distribution, namely Group- A lignment and Global- U niformity Enhanced R epresentation L earning for Debiasing Recommendation (AURL) . Specifically, we identify two significant problems in the representation distribution of users and items, namely group-discrepancy and global-collapse. These two problems directly lead to biases in the recommendation results. To this end, we propose two simple but effective regularizers in the representation space, respectively named group-alignment and global-uniformity. The goal of group-alignment is to bring the representation distribution of long-tail entities closer to that of popular entities, while global-uniformity aims to preserve the information of entities as much as possible by evenly distributing representations. Our method directly optimizes both the group-alignment and global-uniformity regularization terms to mitigate recommendation biases. Please note that AURL applies to arbitrary CF-based recommendation backbones. Extensive experiments on three real datasets and various recommendation backbones verify the superiority of our proposed framework. The results show that AURL not only outperforms existing debiasing models in mitigating biases but also improves recommendation performance to some extent.

Improved Mass Calibration in MALDI MSI Using Neural Network-Based Recalibration.

Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI MSI) is a powerful imaging method for generating molecular maps of biological samples and has numerous applications in biomedical research. A key challenge in MALDI MSI is to reliably map observed mass peaks to theoretical masses, which can be difficult due to mass shifts that occur during the measurement process. In this paper, we propose MassShiftNet, a novel self-supervised learning framework for mass recalibration. We train a neural network on a data dependent and specifically augmented training data set to directly estimate and correct the mass shift in the observed spectra. In our evaluation, we show that this method is both able to reduce the absolute mass error and to increase the relative mass alignment between peptide MSI spectra acquired from FFPE-fixated tissue using a MALDI time-of-flight (TOF) instrument.

Transfer learning enables identification of multiple types of RNA modifications using nanopore direct RNA sequencing

Nanopore direct RNA sequencing (DRS) has emerged as a powerful tool for RNA modification identification. However, concurrently detecting multiple types of modifications in a single DRS sample remains a challenge. Here, we develop TandemMod, a transferable deep learning framework capable of detecting multiple types of RNA modifications in single DRS data. To train high-performance TandemMod models, we generate in vitro epitranscriptome datasets from cDNA libraries, containing thousands of transcripts labeled with various types of RNA modifications. We validate the performance of TandemMod on both in vitro transcripts and in vivo human cell lines, confirming its high accuracy for profiling m6A and m5C modification sites. Furthermore, we perform transfer learning for identifying other modifications such as m7G, Ψ, and inosine, significantly reducing training data size and running time without compromising performance. Finally, we apply TandemMod to identify 3 types of RNA modifications in rice grown in different environments, demonstrating its applicability across species and conditions. In summary, we provide a resource with ground-truth labels that can serve as benchmark datasets for nanopore-based modification identification methods, and TandemMod for identifying diverse RNA modifications using a single DRS sample.

Predicting insurance claims through a variety of data mining techniques: facing lots of missing values and moderate class-imbalanced levels

Abstract This paper copes with a real-world classification problem related to the management of claims received in an insurance company. The way to obtain the classifier is not easy due to the high amount of missing values as well as the inherent imbalanced scenario within class labels. Once the data partition has been done, the training set is submitted to an intensive double grid search in order to obtain the most promising type of missing value imputation approach and then a step ahead is done using the best method and it starts the next round of data mining strategies which now falls into data rebalancing umbrella. Again, a grid search from an undersampling and oversampling family with different settings is done taking into account only seen data. The training data obtained after the first grid search are now submitted to the second step according the second grid search in order to get the ready training set for the further classifier training. The main objective of the work is to find the best combination of data mining techniques that suits the data set with a pipeline containing two types of data preparation methods coming from different families. As an outcome, first the problem of the presence of missing values has been addressed and then the data rebalancing techniques has been applied. The study focuses on obtaining classifiers based on Bayesian and lazy approaches as well as decision trees, evaluated on metrics such as the area under the ROC curve (AUC), Cohen’s kappa, Accuracy and the F-measure, among others. The imputation by the mean the mode is preferable to the Expectation Maximization Imputation in the scenario faced in this paper taking into account that the amount of missing values is higher than a forty percent for many features.

Reducing Training Data Needs with Minimal Multilevel Machine Learning (M3L)

Abstract For many machine learning applications in science, data acquisition, not training, is the bottleneck even when avoiding experiments and relying on computation and simulation. 
Correspondingly, and in order to reduce cost and carbon footprint, training data efficiency is key. 
We introduce minimal multilevel machine learning (M3L) which 
optimizes training data set sizes using a loss function at multiple levels of reference data in order to minimize a combination of prediction error with overall training data acquisition costs (as measured by computational wall-times).
Numerical evidence has been obtained for calculated atomization energies and electron affinities of thousands of organic molecules at various levels of theory including HF, MP2, DLPNO-CCSD(T), DFHFCABS, PNOMP2F12, and PNOCCSD(T)F12, and treating them with 
basis sets TZ, cc-pVTZ, and AVTZ-F12.
Our M3L benchmarks for reaching chemical accuracy in distinct chemical compound sub-spaces 
indicate substantial computational cost reductions by factors of 
$\sim$ 1.01, 1.1, 3.8, 13.8, and 25.8 when compared to heuristic sub-optimal multilevel machine learning (M2L) for the data sets QM7b, QM9$^\mathrm{LCCSD(T)}$, EGP, QM9$^\mathrm{CCSD(T)}_\mathrm{AE}$, and QM9$^\mathrm{CCSD(T)}_\mathrm{EA}$, respectively.
Furthermore, we use M2L to investigate the performance for 76 density functionals when used within multilevel learning and building on the following levels drawn from the hierarchy of Jacobs Ladder:~LDA, GGA, mGGA, and hybrid functionals.
Within M2L and the molecules considered, 
mGGAs do not provide any noticeable advantage over GGAs. 
Among the functionals considered and in combination with LDA, the three on average top performing GGA and Hybrid levels for atomization energies on QM9 using M3L correspond respectively to PW91, KT2, B97D, and $\tau$-HCTH, B3LYP$\ast$(VWN5), and TPSSH.

Machine Learning Algorithms That Emulate Controllers Based on Particle Swarm Optimization—An Application to a Photobioreactor for Algal Growth

Particle Swarm Optimization (PSO) algorithms within control structures are a realistic approach; their task is often to predict the optimal control values working with a process model (PM). Owing to numerous numerical integrations of the PM, there is a big computational effort that leads to a large controller execution time. The main motivation of this work is to decrease the computational effort and, consequently, the controller execution time. This paper proposes to replace the PSO predictor with a machine learning model that has “learned” the quasi-optimal behavior of the couple (PSO and PM); the training data are obtained through closed-loop simulations over the control horizon. The new controller should preserve the process’s quasi-optimal control. In identical conditions, the process evolutions must also be quasi-optimal. The multiple linear regression and the regression neural networks were considered the predicting models. This paper first proposes algorithms for collecting and aggregating data sets for the learning process. Algorithms for constructing the machine learning models and implementing the controllers and closed-loop simulations are also proposed. The simulations prove that the two machine learning predictors have learned the PSO predictor’s behavior, such that the process evolves almost identically. The resulting controllers’ execution time have decreased hundreds of times while keeping their optimality; the performance index has even slightly increased.

Automated prediction of isthmus areas in scar-related atrial tachycardias using artificial intelligence.

Ablation of scar-related reentrant atrial tachycardia (SRRAT) involves identification and ablation of a critical isthmus. A graph convolutional network (GCN) is a machine learning structure that is well-suited to analyze the irregularly-structured data obtained in mapping procedures and may be used to identify potential isthmuses. Electroanatomic maps from 29 SRRATs were collected, and custom electrogram features assessing key tissue and wavefront properties were calculated for each point. Isthmuses were labeled off-line. Training data was used to determine the optimal GCN parameters and train the final model. Putative isthmus points were predicted in the training and test populations and grouped into proposed isthmus areas based on density and distance thresholds. The primary outcome was the distance between the centroids of the true and closest proposed isthmus areas. A total of 193 821 points were collected. Thirty isthmuses were detected in 29 tachycardias among 25 patients (median age 65.0, 5 women). The median (IQR) distance between true and the closest proposed isthmus area centroids was 8.2 (3.5, 14.4) mm in the training and 7.3 (2.8, 16.1) mm in the test group. The mean overlap in areas, measured by the Dice coefficient, was 11.5 ± 3.2% in the training group and 13.9 ± 4.6% in the test group. A GCN can be trained to identify isthmus areas in SRRATs and may help identify critical ablation targets.

Predicting the properties of perovskite materials by improved compositionally restricted attention-based networks and explainable machine learning

Understanding the unique properties of perovskite materials is crucial in advancing solar energy technologies. Factors like heat of formation and bandgap significantly influence the light absorption capability and stability of perovskite solar cells. However, it is time-consuming and labor-intensive to obtain the properties of perovskites using traditional experimental or high-throughput computational methods. As a prospective method, machine learning can find regularities in the given training data and give accurate prediction results. In this article, we use deep learning models based on attention mechanisms and elemental features to predict the heat of formation and bandgap of perovskite materials. Random Forest and Gradient Boosted Regression Tree models have also been used for interpretable predictions of properties. The compositionally restricted attention-based network was improved by introducing a densely connected network and optimizing the network structure to increase data processing capabilities. The experiment results show that the mean absolute errors of the heat of formation and bandgap on the test sets are decreased by 5.77% and 3.37% respectively. The optimized model also shows better performance when used for classification tasks. In addition, we use the gradient boosting regression tree model and the shapley additive explanations tool to conduct an interpretable analysis, explaining the impact of different features on the predictions of the properties.

A novel coupled rainfall prediction model based on stepwise decomposition technique

The traditional decomposed ensemble prediction model decomposes the entire rainfall sequence into several sub-sequences, dividing them into training and testing periods for modeling. During sample construction, future information is erroneously mixed into the training data, making it challenging to apply in practical rainfall forecasting. This paper proposes a novel stepwise decomposed ensemble coupling model, realized through variational mode decomposition (VMD) and bidirectional long short-term memory neural network (BiLSTM) models. Model parameters are optimized using an improved particle swarm optimization (IPSO). The performance of the model was evaluated using rainfall data from the Southern Four Lakes basin. The results indicate that: (1) Compared to the PSO algorithm, the IPSO algorithm-coupled model shows a minimum decrease of 2.70% in MAE and at least 2.62% in RMSE across the four cities in the Southern Four Lakes basin; the IPSO algorithm results in a minimum decrease of 25.58% in MAE and at least 28.19% in RMSE for the VMD-BiLSTM model. (2) When compared to IPSO-BiLSTM, the VMD-IPSO-BiLSTM based on the stepwise decomposition technique exhibits a minimum decrease of 26.54% in MAE and at least 34.16% in RMSE. (3) The NSE for the testing period of the VMD-IPSO-BiLSTM model in each city surpasses 0.88, indicating higher prediction accuracy and providing new insights for optimizing rainfall forecasting.

Development of a ReaxFF reactive force field for ternary phosphate-based bioactive glasses.

Phosphate-based glasses (PBGs) in the CaO-Na2O-P2O5 system have diverse applications as biomaterials due to their unique dissolution properties. A reactive force field (ReaxFF) has been developed to simulate these materials at the atomic level. The ReaxFF parameters of PBGs, including the interaction between phosphorus and calcium atoms, have been developed using a published code based on genetic algorithms. The training data, including the atomic charges, atomic forces, bond and angle parameters, and different differential energies, are chosen and measured from static quantum-mechanical calculations and abinitio molecular dynamics of PBGs. We did different short- and medium-range structural analyses on the bulk simulated PBGs with different compositions to validate the developed potential. Radial and angular distribution functions and coordination numbers of network formers and modifiers, as well as the network connectivity of the glass, are in agreement with experimental and previous simulations using both shell-model classical force fields and abinitio simulated data; for example, the coordination number of phosphorus is 4.0. This successful development of ReaxFF parameters being able to describe the bulk PBGs enables us to work on the dissolution behavior of the glasses, including the interaction of water molecules with PBGs, in future works.

Resolution enhancement for a seismic velocity model using machine learning

Summary To address complex subsurface structures, a high-resolution velocity model must be constructed. Conventionally, algorithms such as full waveform inversion (FWI) have been used to derive accurate high-resolution velocity structures, but obstacles such as high computational costs remain. Therefore, we propose a high-resolution U-NET (HR U-NET) machine learning model to derive a high-resolution velocity model from a low-resolution velocity model. The low-resolution velocity model and migration data obtained through the corresponding velocity information were used as input data for training. In addition, we tried to improve the accuracy of the high-resolution velocity model by using prior information containing accurate velocity values. A prior model generated through geophysical logging data and a weight model including the reliability information of the prior model were also utilized. Therefore, the HR U-NET model was trained using the low-resolution velocity model, the migration data, the prior model, and the weight model. Numerical experiments conducted using synthetic and field data demonstrated that the proposed model could accurately construct a high-resolution velocity model and verified that the prior model and weight model play important roles in the training process. Additionally, we confirmed that the proposed method derived almost similar results using only 8.2 per cent of the computational cost of the conventional inversion method. In other words, there is an advantage that it is possible to predict high-resolution velocity information more efficiently in terms of computational cost.

Detecting Coseismic Landslides in GEE Using Machine Learning Algorithms on Combined Optical and Radar Imagery

Landslides, resulting from disturbances in slope equilibrium, pose a significant threat to landscapes, infrastructure, and human life. Triggered by factors such as intense precipitation, seismic activities, or volcanic eruptions, these events can cause extensive damage and endanger nearby communities. A comprehensive understanding of landslide characteristics, including spatio-temporal patterns, dimensions, and morphology, is vital for effective landslide disaster management. Existing remote sensing approaches mostly use either optical or synthetic aperture radar sensors. Integrating information from both these types of sensors promises greater accuracy for identifying and locating landslides. This study proposes a novel approach, the ML-LaDeCORsat (Machine Learning-based coseismic Landslide Detection using Combined Optical and Radar Satellite Imagery), that integrates freely available Sentinel-1, Palsar-2, and Sentinel-2 imagery data in Google Earth Engine (GEE). The approach also integrates relevant spectral indices and suitable bands used in a machine learning-based classification of coseismic landslides. The approach includes a robust and reproducible training and validation strategy and allows one to choose between five classifiers (CART, Random Forest, GTB, SVM, and Naive Bayes). Using landslides from four different earthquake case studies, we demonstrate the superiority of our approach over existing solutions in coseismic landslide identification and localization, providing a GTB-based detection accuracy of 87–92%. ML-LaDeCORsat can be adapted to other landslide events (GEE script is provided). Transfer learning experiments proved that our model can be applied to other coseismic landslide events without the need for additional training data. Our novel approach therefore facilitates quick and reliable identification of coseismic landslides, highlighting its potential to contribute towards more effective disaster management.

NGMD: next generation malware detection in federated server with deep neural network model for autonomous networks

Distributed denial-of-service (DDoS) attacks persistently proliferate, impacting individuals and Internet Service Providers (ISPs). Deep learning (DL) models are paving the way to address these challenges and the dynamic nature of potential threats. Traditional detection systems, relying on signature-based techniques, are susceptible to next-generation malware. Integrating DL approaches in cloud-edge/federated servers enhances the resilience of these systems. In the Internet of Things (IoT) and autonomous networks, DL, particularly federated learning, has gained prominence for attack detection. Unlike conventional models (centralized and localized DL), federated learning does not require access to users’ private data for attack detection. This approach is gaining much interest in academia and industry due to its deployment on local and global cloud-edge models. Recent advancements in DL enable training a quality cloud-edge model across various users (collaborators) without exchanging personal information. Federated learning, emphasizing privacy preservation at the cloud-edge terminal, holds significant potential for facilitating privacy-aware learning among collaborators. This paper addresses: (1) The deployment of an optimized deep neural network for network traffic classification. (2) The coordination of federated server model parameters with training across devices in IoT domains. A federated flowchart is proposed for training and aggregating local model updates. (3) The generation of a global model at the cloud-edge terminal after multiple rounds between domains and servers. (4) Experimental validation on the BoT-IoT dataset demonstrates that the federated learning model can reliably detect attacks with efficient classification, privacy, and confidentiality. Additionally, it requires minimal memory space for storing training data, resulting in minimal network delay. Consequently, the proposed framework outperforms both centralized and localized DL models, achieving superior performance.

Simulation-Driven Universal Surrogates of Coupled Mechanical Systems: Real-Time Simulation of a Forestry Crane

Abstract Preparing simulation-driven surrogates for a coupled mechanical system can be challenging because the associated mechanical and actuator dynamics demand high-fidelity numerical solutions. Proposed here is a universal hydraulic surrogate (UHS), which can provide solutions to high-fidelity mechanical systems with a universal actuator in a surrogate-assisted monolithic approach. The UHS acts as an alternative to the standard lumped fluid theory by eliminating the hydraulic pressures differential equations. A surrogate-assisted universal actuator uses an approximated model to define hydraulic force in high-fidelity mechanical systems. The approximated force model was developed through training against the dynamics of a one-dimensional (1D) hydraulic cylinder and spring-damper. A covariance matrix adaption evolutionary strategy (CMA-ES) was used as an optimization algorithm to minimize differences between the standard dynamics and UHS approaches at the position and velocity levels. The robustness of resulting UHS was validated to predict the behaviors of the simple four-bar mechanism and the forestry crane. The focus was on numerical accuracy and computational efficiency. The maximum percent normalized root mean square error (PN-RMSE) between the states of the approximated force model and lumped fluid theory were approximately 2.04% and 6.95%, respectively. The proposed method was approximately 52 times faster than the standard lumped fluid theory method. By providing accurate predictions outside the training data, the simulation-driven UHS promises better computational performance leading to real-time simulation solutions for the coupled mechanical systems. The UHS can be applied in simulation, optimization, control, state and parameter estimation, and Artificial Intelligence (AI) implementations for coupled mechanical systems.

Predicting Serotonin Detection with DNA-Carbon Nanotube Sensors across Multiple Spectral Wavelengths.

Owing to the value of DNA-wrapped single-walled carbon nanotube (SWNT)-based sensors for chemically specific imaging in biology, we explore machine learning (ML) predictions DNA-SWNT serotonin sensor responsivity as a function of DNA sequence based on the whole SWNT fluorescence spectra. Our analysis reveals the crucial role of DNA sequence in the binding modes of DNA-SWNTs to serotonin, with a smaller influence of SWNT chirality. Regression ML models trained on existing data sets predict the change in the fluorescence emission in response to serotonin, ΔF/F, at over a hundred wavelengths for new DNA-SWNT conjugates, successfully identifying some high- and low-response DNA sequences. Despite successful predictions, we also show that the finite size of the training data set leads to limitations on prediction accuracy. Nevertheless, incorporating entire spectra into ML models enhances prediction robustness and facilitates the discovery of novel DNA-SWNT sensors. Our approaches show promise for identifying new chemical systems with specific sensing response characteristics, marking a valuable advancement in DNA-based system discovery.

Impact evaluation of a cash-plus programme for children with disabilities in the Xiengkhouang Province in Lao PDR: study protocol for a non-randomised controlled trial

IntroductionMore than 170 countries have implemented disability-targeted social protection programmes, although few have been rigorously evaluated. Consequently, a non-randomised controlled trial is being conducted of a pilot ‘cash-plus’ programme implemented...

Transferring Black-Box Decision Making to a White-Box Model

In the rapidly evolving realm of artificial intelligence (AI), black-box algorithms have exhibited outstanding performance. However, their opaque nature poses challenges in fields like medicine, where the clarity of the decision-making processes is crucial for ensuring trust. Addressing this need, the study aimed to augment these algorithms with explainable AI (XAI) features to enhance transparency. A novel approach was employed, contrasting the decision-making patterns of black-box and white-box models. Where discrepancies were noted, training data were refined to align a white-box model’s decisions closer to its black-box counterpart. Testing this methodology on three distinct medical datasets revealed consistent correlations between the adapted white-box models and their black-box analogs. Notably, integrating this strategy with established methods like local interpretable model-agnostic explanations (LIMEs) and SHapley Additive exPlanations (SHAPs) further enhanced transparency, underscoring the potential value of decision trees as a favored white-box algorithm in medicine due to its inherent explanatory capabilities. The findings highlight a promising path for the integration of the performance of black-box algorithms with the necessity for transparency in critical decision-making domains.

Perbandingan Metode Random Forest, Convolutional Neural Network, dan Support Vector Machine Untuk Klasifikasi Jenis Mangga

Mango is a fruit known as the "King of Fruit" due to its rich flavor, vast variability, and high nutritional value. Classifying mangoes based on their external appearance is the initial step in the process of identifying and categorizing mango types conventionally. The classification process can be performed by examining external features such as fruit color, shape, and size. Classifying different types of mango fruits accurately can assist researchers in developing superior varieties and also aid farmers for cultivation purposes, sales, distribution, and selecting the right varieties for local growth and weather conditions. This research conducts the classification of mango types based on color from mango images using machine learning. The study compares three methods, namely Random Forest, Support Vector Machine (SVM), and Convolutional Neural Network (CNN), to determine the best method for classifying mango types based on their images. The dataset underwent preprocessing, where image sizes were standardized to 300 x 300 pixels, and color was changed to grayscale. The dataset was then divided into training and testing data with a ratio of 70:30. Subsequently, the dataset was processed using three methods, and their accuracy results were compared. The findings indicate that the Random Forest method yielded the highest accuracy compared to the other methods, with an accuracy rate of 96%. The accuracy of the SVM method was 95%, and the accuracy of the CNN method was 33%. From these results, it can be concluded that the Random Forest method is highly effective for classifying mango types based on their image compared to SVM and CNN methods.

Modeling of ordinal longitudinal accelerometer data

Improvements in accelerometer technology has led to new types of data on which more powerful predictive models can be built to assess physical activity. This paper implements an ordinal random forest model with recursive forecasting to take into account the ordinal, longitudinal nature of responses. The data comes from 28 adults performing activities of daily living in two visits, while wearing accelerometers on the ankle, hip, right and left wrist. The first visit provided training data and the second testing data so that an indepen- dent sample, cross-validation approach could be used. For this type of data, prior responses are not available at the testing stage or in practice. However, recursive forecasts can be made with prior predictions in place of lagging responses on models which were built to use lagging responses as explanatory variables. Models are fit to account for multiple time series, with different time series for each participant in the study. We found that ordinal random forest, when the time series is taken into account, produces better accuracy rates and better linearly weighted kappa values than both ordinary ordinal forest and random forest. On the testing set, the lowest error rates were produced by the ankle (28.0%), followed by the left wrist (28.7%), hip (28.9%) and then the right wrist (30.2%) using the best performing decision model for a four-activity level response. In addition, linearly weighted kappa values indicated substantial agreement. The approach of this work can be adapted to other types of longitudinal ordinal models for improvements in modeling techniques. Journal of Statistical Research 2022, Vol. 57, No. 1-2, pp.1-19

Image Fusion Method Based on Snake Visual Imaging Mechanism and PCNN

The process of image fusion is the process of enriching an image and improving the image’s quality, so as to facilitate the subsequent image processing and analysis. With the increasing importance of image fusion technology, the fusion of infrared and visible images has received extensive attention. In today’s deep learning environment, deep learning is widely used in the field of image fusion. However, in some applications, it is not possible to obtain a large amount of training data. Because some special organs of snakes can receive and process infrared information and visible information, the fusion method of infrared and visible light to simulate the visual mechanism of snakes came into being. Therefore, this paper takes into account the perspective of visual bionics to achieve image fusion; such methods do not need to obtain a significant amount of training data. However, most of the fusion methods for simulating snakes face the problem of unclear details, so this paper combines this method with a pulse coupled neural network (PCNN). By studying two receptive field models of retinal nerve cells, six dual-mode cell imaging mechanisms of rattlesnakes and their mathematical models and the PCNN model, an improved fusion method of infrared and visible images was proposed. For the proposed fusion method, eleven groups of source images were used, and three non-reference image quality evaluation indexes were compared with seven other fusion methods. The experimental results show that the improved algorithm proposed in this paper is better overall than the comparison method for the three evaluation indexes.