Optimal Model Selection Research Articles

Optimized Model Selection for Concurrent Decoding of Finger Kinetics and Kinematics

Myoelectric-based motor intent detection is typically used to interface with assistive devices. However, the intent detection performance is sensitive to interference of electromyogram (EMG) signals. Recently, EMG signals are decomposed into motor units (MU) firing activities, and neuron binary firing activities can be used to predict motor output in a continuous manner. Different functions that map MU firings to motor output have been implemented, and both composite MU firing frequency and individual MU firing frequency have been used. It is unclear whether one mapping function outperform others. Accordingly, we evaluated three MU-based finger kinetic and kinematic prediction models, by varying the number of MUs and the method of including MU firings into the regression model. We also compared the performance of three EMG amplitude-based models with varying number of channels. We performed MU decomposition in advance for real-time implementations. Our results showed that individual firing frequency of five MUs provided the lowest estimation error (force: 4.66±0.36 %MVC; joint angle: 4.81±0.49°) and highest correlation (force: 0.86±0.01; joint angle: 0.93±0.01) with the measured motor outputs, when compared with mapping method using the populational firing frequency of all MUs or the populational firing frequency of a group of MUs with similar firing activities. The results indicated that firing information at the population level may mask critical information of individual MU firings. These findings allowed us to identify the optimal models for concurrent and continuous finger force and joint angle estimation. A combination of the minimal level of complexity and high accuracy make these models suitable for real-time control of assistive robotic devices.

Optimal Cluster Head Selection in WSN with Convolutional Neural Network-Based Energy Level Prediction

Currently, analysts in a variety of nations have developed various WSN clustering protocols. The major characteristic is the Low Energy Adaptive Clustering Hierarchy (LEACH), which attained the objective of energy balance by sporadically varying the Cluster Heads (CHs) in the region. Nevertheless, because it implements an arbitrary number system, the appropriateness of CH is complete with suspicions. In this paper, an optimal cluster head selection (CHS) model is developed regarding secure and energy-aware routing in the Wireless Sensor Network (WSN). Here, optimal CH is preferred based on distance, energy, security (risk probability), delay, trust evaluation (direct and indirect trust), and Received Signal Strength Indicator (RSSI). Here, the energy level is predicted using an improved Deep Convolutional Neural Network (DCNN). To choose the finest CH in WSN, Bald Eagle Assisted SSA (BEA-SSA) is employed in this work. Finally, the results authenticate the effectiveness of BEA-SSA linked to trust, RSSI, security, etc. The Packet Delivery Ratio (PDR) for 100 nodes is 0.98 at 500 rounds, which is high when compared to Grey Wolf Optimization (GWO), Multi-Objective Fractional Particle Lion Algorithm (MOFPL), Sparrow Search Algorithm (SSA), Bald Eagle Search optimization (BES), Rider Optimization (ROA), Hunger Games Search (HGS), Shark Smell Optimization (SSO), Rider-Cat Swarm Optimization (RCSO), and Firefly Cyclic Randomization (FCR) methods.

Gaussian Mutation–Spider Monkey Optimization (GM-SMO) Model for Remote Sensing Scene Classification

Scene classification aims to classify various objects and land use classes such as farms, highways, rivers, and airplanes in the remote sensing images. In recent times, the Convolutional Neural Network (CNN) based models have been widely applied in scene classification, due to their efficiency in feature representation. The CNN based models have the limitation of overfitting problems, due to the generation of more features in the convolutional layer and imbalanced data problems. This study proposed Gaussian Mutation–Spider Monkey Optimization (GM-SMO) model for feature selection to solve overfitting and imbalanced data problems in scene classification. The Gaussian mutation changes the position of the solution after exploration to increase the exploitation in feature selection. The GM-SMO model maintains better tradeoff between exploration and exploitation to select relevant features for superior classification. The GM-SMO model selects unique features to overcome overfitting and imbalanced data problems. In this manuscript, the Generative Adversarial Network (GAN) is used for generating the augmented images, and the AlexNet and Visual Geometry Group (VGG) 19 models are applied to extract the features from the augmented images. Then, the GM-SMO model selects unique features, which are given to the Long Short-Term Memory (LSTM) network for classification. In the resulting phase, the GM-SMO model achieves 99.46% of accuracy, where the existing transformer-CNN has achieved only 98.76% on the UCM dataset.

Exploration of the optimal modularity in assembly line design

It is widely accepted that a proper structural modularity degree of assembly processes in terms of mass customization has a positive effect on their efficiency because it, among other things, increases manufacturing flexibility and productivity. On the other hand, most practical approaches to identify such a degree is rather based on intuition or analytical reasoning than on scientific foundations. However, the first way can be used for simple assembly tasks, but in more complex assembly processes, this method lags behind the second. The purpose was to create a methodology for selection of optimal modular assembly model from among a predefined set of alternatives. The methodology is based on exploration of the relations between modularity measures and complexity issues as well as the relationship between structural modularity and symmetry. Especially, the linkage between modularity and complexity properties has been explored in order to show how modularization can affect distribution of the total structural complexity across the entire assembly line. To solve this selection problem, three different methods are preliminary suggested and compared via a series of numerical tests. The two of them present the novel contribution of this work, while the third method developed earlier for the purpose of finding and evaluating community structure in networks was adapted for a given application domain. Based on obtained results, one of these method is prioritized over another, since it offers more promising results and precision too.

Optimal Security Protection Strategy Selection Model Based on Q-Learning Particle Swarm Optimization.

With the rapid development of Industrial Internet of Things technology, the industrial control system (ICS) faces more and more security threats, which may lead to serious risks and extensive damage. Naturally, it is particularly important to construct efficient, robust, and low-cost protection strategies for ICS. However, how to construct an objective function of optimal security protection strategy considering both the security risk and protection cost, and to find the optimal solution, are all significant challenges. In this paper, we propose an optimal security protection strategy selection model and develop an optimization framework based on Q-Learning particle swarm optimization (QLPSO). The model performs security risk assessment of ICS by introducing the protection strategy into the Bayesian attack graph. The QLPSO adopts the Q-Learning to improve the local optimum, insufficient diversity, and low precision of the PSO algorithm. Simulations are performed on a water distribution ICS, and the results verify the validity and feasibility of our proposed model and the QLPSO algorithm.

Effect of learning parameters on the performance of the U-Net architecture for cell nuclei segmentation from microscopic cell images.

Nuclei segmentation of cells is the preliminary and essential step of pathological image analysis. However, robust and accurate cell nuclei segmentation is challenging due to the enormous variability of staining, cell sizes, morphologies, cell adhesion or overlapping of the nucleus. The automation process to find the cell's nuclei is a giant leap in this direction and has an important step toward bioimage analysis using software tools. This article extensively analyzes deep U-Net architecture and has been applied to the Data Science Bowl dataset to segment the cell nuclei. The dataset undergoes various preprocessing tasks such as resizing, intensity normalization and data augmentation prior to segmentation. The complete dataset then undergoes the rigorous training and validation process to find the optimized hyperparameters and then the optimized model selection. The mean (m) ± standard deviation (SD) of Intersection over Union (IoU) and F1-score (Dice score) have been calculated along with accuracy during the training and validation process, respectively. The optimized U-Net model results in a training IoU of 0.94 ± 0.16 (m ± SD), an F1-score of 0.94 ± 0.17 (m ± SD), a training accuracy of 95.54 and validation accuracy of 95.45. With this model, we applied a completely independent test cohort of the dataset and obtained the mean IOU of 0.93, F1-score of 0.9311, and mean accuracy of 94.12, respectively to measure the segmentation performance.

New Dimension on Quality of Life Differences among Older Adults: A Comparative Analysis of Digital Consumption in Urban and Rural Areas of China.

The purpose of this study was to compare the variability in quality of life in the area of digital consumption among older adults in urban and rural China during the COVID-19 pandemic. This study proposed a low-cost mixed research method, and the methodology used a quantitative study of a large regional sample combined with a qualitative study of a small regional sample. Data for the large-scale area sample were obtained from the China family panel study (CFPS) dataset, and data for the small-scale regional sample were obtained from Nanjing, China. The quantitative analysis of the large-scale regional sample used the least squares regression analysis (OSL) and propensity score matching (PSM). The qualitative analysis of the small-scale regional sample used the selection optimization and compensation (SOC) model. The findings show that economic income is a direct driver of digital consumption. Digital consumption had a significant positive relationship with the quality of life for urban and rural older adults. In addition, the study established the semantic network relationships of the coping strategies of digital consumption of older adults and their drivers. Finally, the theoretical and practical implications of these findings are discussed in the context of other related studies.

Real Pandemic World Results of Personalized Prophylaxis of Bleeds in Polish Children and Adolescents with Severe Hemophilia a

Introduction and aim of the study: In the autumn of 2020, the new nationwide protocol of bleeding prophylaxis in Polish plasma-derived FVIII (pdFVIII) previously treated patients (PTPs) with severe hemophilia A (sHA) was introduced, resulting in the necessity of switching from pdFVIII to recombinant FVIII (octocog-alpha; rFVIII). It created the opportunity to perform the study aiming to: 1. assess the safety of switching from pdFVIII to rFVIII 2. assess the safety and efficacy of pharmacokinetically-based (PK-based) personalized prophylaxis in children and adolescents with severe hemophilia A. Patients and methods: A total of one hundred fifty-one patients (median age 166 [93-209] months) receiving prophylaxis with a standard dose (40 U/kg 3 x weekly) of pdFVIII were included in this study. Annualized bleeding rate (ABR) and annualized joint bleeding rate (AJBR) were analyzed for all patients before enrollment. Using myPKFiT application, pharmacokinetic (PK) analysis followed by the selection of optimal (dosing and timing) model of prophylaxis was performed in all patients. All pros and cons of two possible models of prophylaxis (standard-dose rFVIII versus PK-based rFVIII) were discussed with parents/guardians and/or patients leaving the choice to their decision. Episodes of bleeds in two study groups were reported by parents/guardians using e-diaries and verified by local hemophilia treaters. Screening for inhibitor (INH) development was performed every three months and in every case of clinical suspicion. ABR and AJBR were prospectively analyzed again after a minimum follow-up time of 26 weeks with respect to models of prophylaxis and a type of FVIII used. Rates of bleedings were compared using the Student's t-test. Results: Due to COVID-19 pandemic-related problems, 141/151 (93.4%) patients completed the study. Thirty-four patients decided to continue standard prophylaxis with rFVIII (Group I), whereas one hundred seven were switched to PK-based prophylaxis (Group II). The risk of INH development could be assessed in 137/151 (90.7%) patients. Only 2/137 (1.47%) patients (both on PK-based prophylaxis) developed low-titer inhibitors (0.8 BU/ml after 120 exposure days (EDs) and 1.8 BU/ml after 78 EDs, respectively) with no clinical manifestation and with spontaneous elimination of the inhibitor in both cases. The retrospective analysis of bleeds during last 12 months of standard pdFVIII prophylaxis revealed that patients who decided to continue standard prophylaxis (Group I) had better ABR and AJBR than those who started PK-based personalized prophylaxis (Group II). After a minimum of 26 weeks, ABR and AJBR improved significantly in both groups. There was no significant difference in ABR and AJBR between Group I and Group II during the follow-up period, however the rate of reduction of ABR and AJBR was higher in patients on PK-based personalized prophylaxis (Group II). The prophylaxis results are presented in Figure 1. Conclusions: 1. Switching from pdFVIII to rFVIII (octocog-alpha) in PTPs with sHA is safe, 2. PK-based personalized prophylaxis may decrease ABR and AJBR in children and adolescents with sHA, especially in those in whom results of previous prophylaxis were unsatisfactory. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Goal programming and genetic algorithm in multiple objective optimization model for project portfolio selection: a review

Many scientific fields, such as engineering, data analytics, and deep learning, focus on optimization. Optimization problems are classified into two types based on the number of optimized objective functions: single objective and multi objective optimization problems. In this paper, a comparative review since 2000 using one of the deterministic and stochastic modelling approaches called goal programming (GP) and genetic algorithm (GA) in multi-objective optimization problem is discussed. This study gives a prime review of the application of GP and GA in various criteria of project portfolio selection problem. GP is a method for solving large-scale multi-objective optimization problems to assist decision makers in finding solutions that satisfy several competing goals. GA on the other hand are global meta-heuristic search algorithms that are used to provide approximation or optimal solutions to large-scale optimization problems. Of the 23 articles considered in this review showed that, from more than 100 projects, GA proved near optimal, feasible solution and efficient frontier in projects ranking, projects interaction and a preferred decision support tool of project portfolio selection. In addition, the two models select projects on risk-based approach, but GA proved to be more effective in terms of number of projects proposed, central processing unit (CPU) time and accuracy. The review concludes that, in multi-objective optimization model for project portfolio selection problems on a large-scale, very large or complex problems and less CPU time, GA is more effective than GP in multi-objective optimization problems. The review also showed gaps in previous studies of GP and GA application on project portfolio selection problem (PPSP). This review will aid scholars and demanding practitioners in gaining a broader understanding of goal programming and genetic algorithms in the context of project portfolio selection problems.

Pairwise-Comparison Based Semi-SPO Method for Ship Inspection Planning in Maritime Transportation

Port state control (PSC) plays an important role in enhancing maritime safety and protecting the marine environment. Since the inspection resources are limited and the inspection process is costly and time-consuming, a critical issue for port states to guarantee inspection efficiency is to accurately select ships with a high risk for inspection. To address this issue, this study proposes three prediction models to predict the ship deficiency number and a ship selection optimization model based on the prediction results to target the riskiest ships for inspection. In addition to a linear regression model for ship deficiency number prediction solved by the least squares method, we establish two prediction models with the pairwise-comparison target based semi-“smart predict then optimize” (semi-SPO) method. Specifically, a linear programming model and a support vector machine (SVM) model are built and both have a loss function to minimize the sum of predicted ranking errors of each pair of ships regarding their deficiency numbers. We use the Hong Kong port as a case study, which shows that the SVM model based on the semi-SPO approach performs best among the three models with the least computation time and best ship selection decisions.

Taming the Selection of Optimal Substitution Models in Phylogenomics by Site Subsampling and Upsampling.

The selection of the optimal substitution model of molecular evolution imposes a high computational burden for long sequence alignments in phylogenomics. We discovered that the analysis of multiple tiny subsamples of site patterns from a full sequence alignment recovers the correct optimal substitution model when sites in the subsample are upsampled to match the total number of sites in the full alignment. The computational costs of maximum-likelihood analyses are reduced by orders of magnitude in the subsample-upsample (SU) approach because the upsampled alignment contains only a small fraction of all site patterns. We present an adaptive protocol, ModelTamer, that implements the new SU approach and automatically selects subsamples to estimate optimal models reliably. ModelTamer selects models hundreds to thousands of times faster than the full data analysis while needing megabytes rather than gigabytes of computer memory.

Selective maintenance optimization with stochastic break duration based on reinforcement learning

For industrial and military applications, a sequence of missions would be performed with a limited break between two adjacent missions. To improve the system reliability, selective maintenance may be performed on components during the break. Most studies on selective maintenance generally use minimal repair and replacement as maintenance actions while break duration is assumed to be deterministic. However, in practical engineering, many maintenance actions are imperfect maintenance, and the break duration is stochastic due to environmental and other factors. Therefore, a selective maintenance optimization model is proposed with imperfect maintenance for stochastic break duration. The model is aimed to maximize the reliability of system successfully completing the next mission. The reinforcement learning(RL) method is applied to optimally select maintenance actions for selected components. The proposed model and the advantages of the RL are verified by three case studies verify.

Genetic Algorithm Based Model for Optimal Selection of Open Channel Design Parameters

Open channels are one of the most used water conveyance systems for delivering water for different purposes. Existing models for the design of open channels mainly assume uniform flow, focus on cross-section sizing, and generally decouple cross-section sizing from the selection of channel alignment and profile. In this study, we developed an optimization model for a comprehensive design of transmission channels. The model minimizes the sum of costs for earthwork, lining, water losses, and land acquisition; accounts for non-uniform, mixed-regime flow; and considers multiple geometric and hydraulic constraints. The model was validated using several idealized scenarios. The model potential in minimizing the cost of real open channel projects was demonstrated through application to an existing irrigation water transmission canal in Egypt (the Sheikh Zayed Canal). The results of validation scenarios matched the anticipated outcomes for channel profile and alignment and reproduced analytical solutions given in the literature for channel cross-section design. Application of the model to the Sheikh Zayed Canal gave a more optimal design; the OCCD model produced a design alternative with ~27% less cost than the constructed alternative.

Spatial Evaluation of Machine Learning-Based Species Distribution Models for Prediction of Invasive Ant Species Distribution

Recent advances in species distribution models (SDMs) associated with artificial intelligence (AI) and increased volumes of available data for model variables have allowed reliable evaluation of the potential distribution of any species. A reliable SDM requires suitable occurrence records and variables with optimal model structures. In this study, we developed three different machine learning-based SDMs [MaxEnt, random forest (RF), and multi-layer perceptron (MLP)] to predict the global potential distribution of two invasive ants under current and future climates. These SDMs showed that the potential distribution of Solenopsis invicta would be expanded by climatic change, whereas it would not significantly change for Anoplolepis gracilipes. The models were compared using model performance metrics, and the optimal model structure and spatial projection were selected. The MaxEnt exhibited high performance, while the MLP model exhibited low performance, with the largest variation by climate change. Random forest showed the smallest potential distribution area, but it was robust considering the number of occurrence records and changes in model variables. All the models showed reliable performance, but the difference in performance and projection size suggested that optimal model selection based on data availability, model variables, study objectives, or an ensemble approach was necessary to develop a comprehensive SDM to minimize modeling uncertainty. We expect that this study will help with the use of AI-based SDMs for the evaluation and risk assessment of invasive ant species.

Prognostic utility of a multibiomarker panel in patients with suspected myocardial infarction

Abstract Introduction Rapid and accurate evaluation of patients presenting with suspected myocardial infarction (MI) to the emergency department (ED) is crucial due to strongly varying outcomes in these patients. Despite the routine use of circulating biomarkers, among which high-sensitivity cardiac troponin currently dictates diagnostic protocols in chest pain patients, it still remains unclear which biomarkers are of highest utility for prognostic purposes in this patient collective. Purpose We sought to investigate the prognostic utility of a multibiomarker panel with 29 different biomarkers in a contemporary cohort of patients with suspected MI by applying a dual analytical approach. Methods The multibiomarker panel was measured in stored blood samples that were collected directly at admission from 748 prospectively enrolled patients who presented with symptoms suggestive of MI to the ED of a German tertiary center between 2013 and 2017. The final diagnosis of all patients was adjudicated according to the 4th Universal Definition of MI. The investigated endpoint comprised incident major cardiovascular events (MACE) within 1 year after admission. MACE was defined as the composite of cardiovascular death, non-fatal MI (excluding index events), revascularization and cardiac rehospitalization. Log-transformed biomarkers were entered individually into an age-and sex-adjusted Cox regression model to explore the hazard ratio (HR) per one standard deviation increase (SD) for 1-year MACE in the overall cohort. The selection of optimal multimarker models, adjusted for age and sex, was performed using 1) stepwise backward selection via Akaike information criterion as the stopping rule, and 2) Least Absolute Shrinkage and Selection Operator (LASSO) with 5-fold cross-validation. Results Out of 748 patients with available multibiomarker panel, 138 (18.4%) were diagnosed with MI. Median age at admission was 64 (interquartile range [IQR] 50–75) years in the overall cohort, 63.1% were male. At 1 year of follow-up, 160 cases of incident MACE were documented. 16 of the investigated 29 biomarkers were significantly associated with 1-year MACE, amongst which NT-proBNP was the strongest predictor (HR per SD 1.74, 95% confidence interval [CI] 1.44–2.09, p<0.0001; Table 1). Using stepwise backward selection, three biomarkers including NT-proBNP, Apo A-I and Apo C-III remained in the final multivariable model (Figure 1). The LASSO approach confirmed NT-proBNP (HR per SD 1.24) and Apo A-I (HR per SD 0.98) as strong and independent predictors of 1-year MACE while Apo C-III was replaced with KIM-1 (HR per SD 1.06). Conclusions Among 29 biomarkers, numerous provide prognostic utility while NT-proBNP and Apo A-I emerged as the strongest independent predictors of 1-year MACE. Their routine assessment and integration into risk prediction models may improve personalized risk stratification in patients with symptoms suggestive of MI. Funding Acknowledgement Type of funding sources: Foundation. Main funding source(s): The BACC study was supported by the German Center for Cardiovascular Research, an unrestricted grant by Abbott Diagnostics, and Prevencio, which also partly covered the biomarker measurements.

SHiFT

Transfer learning can be seen as a data- and compute-efficient alternative to training models from scratch. The emergence of rich model repositories, such as TensorFlow Hub, enables practitioners and researchers to unleash the potential of these models across a wide range of downstream tasks. As these repositories keep growing exponentially, efficiently selecting a good model for the task at hand becomes paramount. However, a single generic search strategy (e.g., taking the model with the highest linear classifier accuracy) does not lead to optimal model selection for diverse downstream tasks. In fact, using hybrid or mixed strategies can often be beneficial. Therefore, we propose SHiFT, the first downstream task-aware, flexible, and efficient model search engine for transfer learning. Users interface with SHiFT using the SHiFT-QL query language, which gives users the flexibility to customize their search criteria. We optimize SHiFT-QL queries using a cost-based decision maker and evaluate them on a wide rang of tasks. Motivated by the iterative nature of machine learning development, we further support efficient incremental executions of our queries, which requires a special implementation when jointly used with our optimizations.

Enhanced Heart Disease Prediction Based on Machine Learning and χ2 Statistical Optimal Feature Selection Model

Automatic heart disease prediction is a major global health concern. Effective cardiac treatment requires an accurate heart disease prognosis. Therefore, this paper proposes a new heart disease classification model based on the support vector machine (SVM) algorithm for improved heart disease detection. To increase prediction accuracy, the χ2 statistical optimum feature selection technique was used. The suggested model’s performance was then validated by comparing it to traditional models using several performance measures. The proposed model increased accuracy from 85.29% to 89.7%. Additionally, the componential load was reduced by half. This result indicates that our system outperformed other state-of-the-art methods in predicting heart disease.

Improved Whale Optimization Algorithm Based on Hybrid Strategy and Its Application in Location Selection for Electric Vehicle Charging Stations

The charging station location model is a nonlinear programming model with complex constraints. In order to solve the problems of weak search ability and low solution accuracy of the whale optimization algorithm (WOA) in solving location models or high-dimensional problems, this paper proposes an improved whale optimization algorithm (IWOA) based on hybrid strategies. Chaos mapping and reverse learning mechanism are introduced in the original algorithm, and the change mode of convergence factor and probability threshold is improved. Through optimization experiments on 18 benchmark functions, the test results show that IWOA has the best solution ability. Finally, IWOA is used to solve a site selection optimization model aiming at the minimum comprehensive cost. The results show that the proposed algorithm and model can effectively reduce the comprehensive cost of site selection. This provides a necessary decision-making reference for the scientific site selection for electric vehicle charging stations.

Portfolio Selection Models Based on Interval-Valued Conditional Value-at-Risk (ICVaR) and Case Study on the Data from Stock Markets

Risk management is very important for individual investors or companies. There are several ways to measure the risk of investment. Prices of risky assets vary rapidly and randomly due to the complexity of finance market. Random interval is a good tool to describe uncertainty including both randomness and imprecision. Considering the uncertainty of financial market, we employ random intervals to describe returns of a risk asset and define an interval-valued risk measurement, which considers the tail risk. It is called the interval-valued conditional value-at-risk (ICVaR, for short). Similar to the classical conditional value-at-risk, ICVaR satisfies the sub-additivity. Under the new risk measure ICVaR, as a manner similar to the classical Mean-CVaR portfolio model, two optimal interval-valued portfolio selection models are built. The sub-additivity of ICVaR guarantees the global optimal solution to the Mean-ICVaR portfolio model. Based on the real data from mainland Chinese stock markets and international stock markets, the case study shows that our models are interpretable and consistent with the practical scenarios.

Cost-Based or Learning-Based?

Traditional cost-based optimizers are efficient and stable to generate optimal plans for simple SQL queries, but they may not generate high-quality plans for complicated queries. Thus learning-based optimizers have been proposed recently that can learn high-quality plans based on past experiences. However, learning-based optimizers cannot work well for dynamic workloads that have different distributions with training examples. In this paper, we propose a hybrid optimizer that adopts the advantages and avoids the shortcomings of these two types of optimizers, which first generates high-quality candidate plans from each type of optimizers and then selects the best plan from the candidates. There are two challenges. (1) How to generate high-quality candidates? We propose a hint-based candidate generation method that leverages the learning-based method to generate highly beneficial hints and then uses a cost-based method to supplement the hints to generate complete plans as candidates. (2) How to evaluate different candidate plans and select the best one? We propose an uncertainty-based optimal plan selection model, which predicts the execution time and the uncertainty for each plan. The uncertainty reflects the confidence of the execution time prediction. We select the plan using the uncertainty model. Experiment results on real datasets showed that our method outperformed the state-of-the-art baselines, and reduced the total latency by 25% and the tail latency by 65% compared to PostgreSQL.