• Computers and Concrete
• /
• v.34 no.3
• /
• pp.279-296
• /
• 2024

In this study, ensemble machine learning (ML) models are employed to estimate the hardened properties of Self-Compacting Geopolymer Concrete (SCGC). The input variables affecting model development include the content of the SCGC such as the binder material, the age of the specimen, and the ratio of alkaline solution. On the other hand, the output parameters examined includes compressive strength, flexural strength, and split tensile strength. The ensemble machine learning models are trained and validated using a database comprising 396 records compiled from 132 unique mix trials performed in the laboratory. Diverse machine learning techniques, notably K-nearest neighbours (KNN), Random Forest, and Extreme Gradient Boosting (XGBoost), have been employed to construct the models coupled with Bayesian optimisation (BO) for the purpose of hyperparameter tuning. Furthermore, the application of nested cross-validation has been employed in order to mitigate the risk of overfitting. The findings of this study reveal that the BO-XGBoost hybrid model confirms better predictive accuracy in comparison to other models. The R² values for compressive strength, flexural strength, and split tensile strength are 0.9974, 0.9978, and 0.9937, respectively. Additionally, the BO-XGBoost hybrid model exhibits the lowest RMSE values of 0.8712, 0.0773, and 0.0799 for compressive strength, flexural strength, and split tensile strength, respectively. Furthermore, a SHAP dependency analysis was conducted to ascertain the significance of each parameter. It is observed from this study that GGBS, Flyash, and the age of specimens exhibit a substantial level of influence when predicting the strengths of geopolymers.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.176-176
• /
• 2018

This study presents a regional, probabilistic framework for estimating streamflow via spatial scaling in the Great Lakes basin, which is the largest lake system in the world. The framework follows a two-fold strategy including (1) a quadratic-programming based optimization model a priori to explore the model structure, and (2) a time-varying hierarchical Bayesian model based on insights found in the optimization model. The proposed model is developed to explore three innovations in hierarchical modeling for reconstructing historical streamflow at ungaged sites: (1) information of physical characteristics is utilized in spatial scaling, (2) a time-varying approach is introduced based on climate information, and (3) heteroscedasticity in residual errors is considered to improve streamflow predictive distributions. The proposed model is developed and calibrated in a hierarchical Bayesian framework to pool regional information across sites and enhance regionalization skill. The model is validated in a cross-validation framework along with four simpler nested formulations and the optimization model to confirm specific hypotheses embedded in the full model structure. The nested models assume a similar hierarchical Bayesian structure to our proposed model with their own set of simplifications and omissions. Results suggest that each of three innovations improve historical out-of-sample streamflow reconstructions although these improvements vary corrsponding to each innovation. Finally, we conclude with a discussion of possible model improvements considered by additional model structure and covariates.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.105-127
• /
• 2024

We propose recurrent neural network (RNN) methods for forecasting realized volatility (RV). The data are RVs of ten major stock price indices, four from the US, and six from the EU. Forecasts are made for relative ratio of adjacent RVs instead of the RV itself in order to avoid the out-of-scale issue. Forecasts of RV ratios distribution are first constructed from which those of RVs are computed which are shown to be better than forecasts constructed directly from RV. The apparent asymmetry of RV ratio is addressed by the Piecewise Min-max (PM) normalization. The serial dependence of the ratio data renders us to consider two architectures, long short-term memory (LSTM) and gated recurrent unit (GRU). The hyperparameters of LSTM and GRU are tuned by the nested cross validation. The RNN forecast with the PM normalization and ratio transformation is shown to outperform other forecasts by other RNN models and by benchmarking models of the AR model, the support vector machine (SVM), the deep neural network (DNN), and the convolutional neural network (CNN).

• ETRI Journal
• /
• v.45 no.1
• /
• pp.105-118
• /
• 2023

We tested the feasibility of automated discrimination of patients with panic disorder (PD) from healthy controls (HCs) based on multimodal physiological responses using machine learning. Electrocardiogram (ECG), electrodermal activity (EDA), respiration (RESP), and peripheral temperature (PT) of the participants were measured during three experimental phases: rest, stress, and recovery. Eleven physiological features were extracted from each phase and used as input data. Logistic regression (LoR), k-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), and multilayer perceptron (MLP) algorithms were implemented with nested cross-validation. Linear regression analysis showed that ECG and PT features obtained in the stress and recovery phases were significant predictors of PD. We achieved the highest accuracy (75.61%) with MLP using all 33 features. With the exception of MLP, applying the significant predictors led to a higher accuracy than using 24 ECG features. These results suggest that combining multimodal physiological signals measured during various states of autonomic arousal has the potential to differentiate patients with PD from HCs.

• Journal of Gastric Cancer
• /
• v.22 no.2
• /
• pp.120-134
• /
• 2022

Purpose: This study aimed to identify prognostic factors for patients with distant lymph node-involved gastric cancer (GC) using a machine learning algorithm, a method that offers considerable advantages and new prospects for high-dimensional biomedical data exploration. Materials and Methods: This study employed 79 features of clinical pathology, laboratory tests, and therapeutic details from 289 GC patients whose distant lymphadenopathy was presented as the first episode of recurrence or metastasis. Outcomes were measured as any-cause death events and survival months after distant lymph node metastasis. A prediction model was built based on possible outcome predictors using a random survival forest algorithm and confirmed by 5×5 nested cross-validation. The effects of single variables were interpreted using partial dependence plots. A contour plot was used to visually represent survival prediction based on 2 predictive features. Results: The median survival time of patients with GC with distant nodal metastasis was 9.2 months. The optimal model incorporated the prealbumin level and the prothrombin time (PT), and yielded a prediction error of 0.353. The inclusion of other variables resulted in poorer model performance. Patients with higher serum prealbumin levels or shorter PTs had a significantly better prognosis. The predicted one-year survival rate was stratified and illustrated as a contour plot based on the combined effect the prealbumin level and the PT. Conclusions: Machine learning is useful for identifying the important determinants of cancer survival using high-dimensional datasets. The prealbumin level and the PT on distant lymph node metastasis are the 2 most crucial factors in predicting the subsequent survival time of advanced GC.