K-fold Approach Research Articles

Machine learning techniques to evaluate the ultrasonic pulse velocity of hybrid fiber-reinforced concrete modified with nano-silica

It is evident that preparing materials, casting samples, curing, and testing all need time and money. The construction sector will benefit if these problems can be handled using cutting-edge techniques like machine learning. Also, a material’s ultrasonic pulse velocity (UPV) is affected by various variables, and it is difficult to study their combined effect experimentally. This research used machine learning to assess the UPV and SHapley Additive ExPlanations techniques to study the impact of input parameters of hybrid fiber-reinforced concrete modified with nano-silica (HFRNSC). Three ML algorithms were employed, i.e., gradient boosting regressor, adaptive boosting regressor, and extreme gradient boosting, for ultrasonic pulse velocity evaluation. The accuracy of machine learning models was measured via the coefficient of determination (R2), k-fold analysis, statistical tests, and comparing the predicted and actual ultrasonic pulse velocity. This study determined that the gradient boosting and adaptive boosting models had a good level of accuracy for ultrasonic pulse velocity, but the extreme gradient boosting method estimated the ultrasonic pulse velocity of HFRNSCs with a greater degree of precision. Also, from the statistical checks and k-fold approach, it was discovered that the extreme gradient boosting method is more exact in estimating the ultrasonic pulse velocity of HFRNSCs. The SHapley Additive ExPlanations analysis revealed that the age of the specimen and nano-silica had a greater positive impact on the ultrasonic pulse velocity of HFRNSCs, whereas the coarse aggregate to fine aggregate ratio had a negative impact. In addition, fiber volume was found to have both positive and negative effects. By aiding the development of rapid and low-cost methods for determining material properties and the influence of input parameters, the construction industry may profit from the use of such technologies.

Probabilistic Forecasts of Flood Inundation Maps Using Surrogate Models

The use of data-driven surrogate models to produce deterministic flood inundation maps in a timely manner has been investigated and proposed as an additional component for flood early warning systems. This study explores the potential of such surrogate models to forecast multiple inundation maps in order to generate probabilistic outputs and assesses the impact of including quantitative precipitation forecasts (QPFs) in the set of predictors. The use of a k-fold approach for training an ensemble of flood inundation surrogate models that replicate the behavior of a physics-based hydraulic model is proposed. The models are used to forecast the inundation maps resulting from three out-of-the-dataset intense rainfall events both using and not using QPFs as a predictor, and the outputs are compared against the maps produced by a physics-based hydrodynamic model. The results show that the k-fold ensemble approach has the potential to capture the uncertainties related to the process of surrogating a hydrodynamic model. Results also indicate that the inclusion of the QPFs has the potential to increase the sharpness, with the tread-off also increasing the bias of the forecasts issued for lead times longer than 2 h.

Implementation of Logistic Regression on Diabetic Dataset using Train-Test-Split, K-Fold and Stratified K-Fold Approach

Implementation of Logistic Regression on Diabetic Dataset using Train-Test-Split, K-Fold and Stratified K-Fold Approach

An Experimental Study to Locate Leakage in the Water Distribution Network using Real-time Wireless Sensor and Machine Learning Algorithm

Objective: To pinpoint the leakage location using a machine learning algorithm on a real-time basis. Method: A laboratory experimental model was developed using wireless sensors for real-time data collection and monitoring the changes in the pressure and flow in presence of leakage under a different scenario. Modification in laboratory model was made considering Loop network of distribution pipes as compared to the simple experimental model. The model has been validated in EPANET software. The machine learning algorithm, along with the K-fold approach has been used to locate leakage and it was compared with other algorithms like Support Vector Machine (SVM), logistic regression, multilayer perceptron, Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM), Principal Component Analysis (PCA), Artificial Neural Network (ANN), Random Forest, Gradient Boosting Tree (GBT) Classification, and Convolutional Neural Network (CNN). Findings: In this research, a system was developed based on real-time data received by Wireless Sensor Network (WSN) was demonstrated and it is able to find even a small leak by monitoring pressure and flow. The K-fold approach in machine learning algorithm has been used to locate the leakage in different variations made in the experimental model in terms of pressure, leak size, cross-section of pipe, and profile level of the pipe network. A comparison was also presented with other machine learning algorithms of recent research in terms of accuracy, size of the opening for the leak, type of experimental model used, and variables considered to locate the leakage. Novelty: It is the first research kind of work that shows the average accuracy of 78% to locate leakage based on the real-time experimental data in a loop network using the K-Fold approach. Keywords: Leakage; Leakage Detection; Kfold approach; Pressure Measurements; Water Distribution Network; Wireless sensor

Kinetic model assessment for the synthesis of γ-valerolactone from n-butyl levulinate and levulinic acid hydrogenation over the synergy effect of dual catalysts Ru/C and Amberlite IR-120

The production of platform molecules from the valorization of lignocellulosic biomass is increasing. Among these plateform molecules, γ-valerolactone (GVL) is a promising one and could be used for different industrial applications. This molecule is synthesized from levulinic acid (LA) or alkyl levulinates (AL) through a tandem hydrogenation/cyclization (lactonization) cascade. A lot of investigations have been carried out to develop the best catalyst for the hydrogenation step by using solely LA or AL. However, one should keep in mind that in the AL production via fructose alcoholysis, there is also LA production, and both are present in the product mixture during the further conversion. To the best of our knowledge, no article exists describing the hydrogenation of LA and AL simultaneously in one-pot. Also, the literature reporting the use of solid catalyst for the second cyclization step is rare. To fill this gap, the hydrogenation of levulinic acid and butyl levulinate (BL) was studied over Ru/C and Amberlite IR-120. Several kinetic models were evaluated via Bayesian inference and K-fold approach. The kinetic assessment showed that a non-competitive Langmuir-Hinshelwood with no dissociation of hydrogen where LA, BL and H2 are adsorbed on different sites (NCLH1.2) and non-competitive Langmuir-Hinshelwood with dissociation of hydrogen where LA, BL and H2 are adsorbed on different sites (NCLH2.2) are the best model to describe this system. The presence of LA and Amberlite IR-120 allows to increase the kinetics of cyclization steps, and in fine to accelerate the production of GVL.

REGULARIZATION METHODS FOR ESTIMATING A MULTI-FACTOR CORPORATE BOND PRICING MODEL: AN APPLICATION FOR BRAZIL

The present work focused on studying which factors affect Brazilian inflation-linked corporate bond prices in a primary market setting. The explanatory variables tested were rating, maturity, duration, issuer governance level, industrial classification, collateral, tax exemption, public offering modality, financial volume, coupon frequency, number of issues, number of days since going public, and the Brazilian basic interest rate target. In order to choose the set of variables with best predictive performance, best subsets ordinary least square (OLS) and least absolute shrinkage and selection operator (LASSO) were applied on a testing sample. For estimating purposes, we also tested the Ridge estimator. For both LASSO and Ridge, we used the k-fold approach to choose the optimal value for the lambda penalty. In terms of smallest mean squared error, the OLS estimator outperformed both the Ridge and the LASSO. This result suggests that the variance-bias trade-off might not be a concern for the Brazilian case.

A Wrapper Feature Selection Algorithm: An Emotional Assessment Using Physiological Recordings from Wearable Sensors.

Assessing emotional state is an emerging application field boosting research activities on the topic of analysis of non-invasive biosignals to find effective markers to accurately determine the emotional state in real-time. Nowadays using wearable sensors, electrocardiogram and thoracic impedance measurements can be recorded, facilitating analyzing cardiac and respiratory functions directly and autonomic nervous system function indirectly. Such analysis allows distinguishing between different emotional states: neutral, sadness, and disgust. This work was specifically focused on the proposal of a k-fold approach for selecting features while training the classifier that reduces the loss of generalization. The performance of the proposed algorithm used as the selection criterion was compared to the commonly used standard error function. The proposed k-fold approach outperforms the conventional method with 4% hit success rate improvement, reaching an accuracy near to 78%. Moreover, the proposed selection criterion method allows the classifier to produce the best performance using a lower number of features at lower computational cost. A reduced number of features reduces the risk of overfitting while a lower computational cost contributes to implementing real-time systems using wearable electronics.

Prediction intervals for global solar irradiation forecasting using regression trees methods

Abstract A global horizontal irradiation prediction (from 1 h to 6 h) is performed using 2 persistence models (simple and “smart” ones) and 4 machine learning tools belonging to the regression trees methods family (normal, pruned, boosted and bagged). A prediction band is associated to each forecast using methodologies based on: bootstrap sampling and k-fold approach, mutual information, stationary time series process with clear sky model, quantiles estimation and cumulative distribution function. New reliability indexes (gamma index and gamma test) are built from the mean interval length (MIL) and prediction interval coverage probability (PCIP). With such methods and error metrics, good prediction bands are estimated for Ajaccio (France) with a MIL close to 113 Wh/m2, a PCIP reaching 70% and a gamma index lower than 0.9.

A learning algorithm for feature weighting in case-based reasoning

Case-based reasoning (CBR) is a methodology for problem-solving and decision making in complex business environment. This methodology is based on the reasoning of earlier experiences to solve new problems. Among well-known methods for case retrieval, K-nearest neighbour (K-NN) is more widely used. But since it is highly sensitive to input features, irrelevant features may cause an increase in the prediction error. In this research, a learning algorithm is developed by applying performance feedback to assign weight setting. The main idea behind the algorithm is to recognise cases which are incorrectly selected as nearest neighbours due to ineffective feature weighting assignment, and attempts to iteratively modify feature weights. This approach is applied to provide suggestions for better decision making and managing strategies of products in the middle of life phase. The dataset of a department store, including various products, have been used. K-fold approach is used to evaluate the prediction accuracy rate. Empirical results are compared with artificial neural network, classical CBR, CBR-GA algorithms. The results indicate that the predictive performance of the model, compare with two CBR-based algorithms, in specific case is more effective.