Boosting Regression Algorithm Research Articles

Interpretable machine learning for simultaneous designing martensitic transformation temperature and thermal hysteresis of high-entropy shape memory alloys

As an excellent functional material, high-entropy shape memory alloy (HESMAs) exhibits tremendous application potential in various fields such as actuators and energy exploration due to its advantages of high strength, high hardness, and corrosion resistance. The key factors determining the application potential of HESMAs are martensitic transformation temperature and thermal hysteresis. However, the design of HESMAs is a great challenge due to the large component search space. Here, we firstly propose a workflow aiming to achieve dual objectives: predicting the martensitic transformation peak temperature (Mp) and thermal hysteresis (Thy) to achieve a rapid design for HESMAs. The key features are obtained by correlation screening and Borutashap feature screening. The Mp and Thy models are training by eXtreme Gradient Boosting Regression algorithm(XGBR) combined with Bayesian Optimization algorith. Finally, XGBR model demonstrates superior performance in predicting both Mp (Rtrain2= 0.993, Rtest2= 0.902) and Thy (Rtrain2= 0.998, Rtest2= 0.928). By applying these models, we are able to identify potential candidates for targeted HESMAs in vast search spaces, specifically, low Mp and wide Thy HESMAs,room Mp and narrow Thy HESMAs, and high Mp HESMA. In addition, we utilize the Shapley additive interpretation method to explain the model's behavior, establishing a more transparent relationship between features and Mp and Thy individually. This proposed strategy is providing valuable guidance for the subsequent targeted design of HESMAs.

A wearable knee rehabilitation system based on graphene textile composite sensor: Implementation and validation

The effectiveness of knee rehabilitation systems in aiding patients with rehabilitation training has been well-documented. Presently, there is an increasing emphasis on the wearing comfort and user engagement of these systems. In this paper, a wearable knee rehabilitation system (WKRS) based on the graphene textile composite sensor (GTCS), which subjects perform rehabilitation training by controlling the ascent and descent of a bird in the visual feedback game utilized GTCS, is proposed and investigated. To obtain an accurate and smooth estimated knee joint angle, we propose an improved tree boosting regression algorithm based on extreme gradient boosting (XGBoost), called improved XGBoost (IXGB). Specifically, the estimated results are smoothed by two steps: curving the preliminary estimating results within the same interval and smoothing the preliminary estimating results using the weighted moving average method. An online experiment with ten subjects validates the effectiveness of the WKRS and IXGB. Results indicate that the IXGB significantly enhances the smoothness of estimated value while maintaining estimation accuracy compared to XGBoost, random forest regression (RFR) and support vector regression (SVR). IXGB achieves an average increase of 33.43 % in root mean square error, 23.83 % in R-squared, and a 62.63% decrease in smoothness.

Machine Learning Prediction of Quantum Yields and Wavelengths of Aggregation-Induced Emission Molecules.

The aggregation-induced emission (AIE) effect exhibits a significant influence on the development of luminescent materials and has made remarkable progress over the past decades. The advancement of high-performance AIE materials requires fast and accurate predictions of their photophysical properties, which is impeded by the inherent limitations of quantum chemical calculations. In this work, we present an accurate machine learning approach for the fast predictions of quantum yields and wavelengths to screen out AIE molecules. A database of about 563 organic luminescent molecules with quantum yields and wavelengths in the monomeric/aggregated states was established. Individual/combined molecular fingerprints were selected and compared elaborately to attain appropriate molecular descriptors. Different machine learning algorithms combined with favorable molecular fingerprints were further screened to achieve more accurate prediction models. The simulation results indicate that combined molecular fingerprints yield more accurate predictions in the aggregated states, and random forest and gradient boosting regression algorithms show the best predictions in quantum yields and wavelengths, respectively. Given the successful applications of machine learning in quantum yields and wavelengths, it is reasonable to anticipate that machine learning can serve as a complementary strategy to traditional experimental/theoretical methods in the investigation of aggregation-induced luminescent molecules to facilitate the discovery of luminescent materials.

Students Final Academic Score Prediction Using Boosting Regression Algorithms

Students Final Academic Score Prediction Using Boosting Regression Algorithms

Machine learning regression algorithms to predict emissions from steam boilers

Currently, the modeling of complex chemical-physical processes is drastically influencing industrial development. Therefore, the analysis and study of the combustion process of the boilers using machine learning (ML) techniques are vital to increase the efficiency with which this equipment operates and reduce the pollution load they contribute to the environment. This work aims to predict the emissions of CO, CO2, NOx, and the temperature of the exhaust gases of industrial boilers from real data. Different ML algorithms for regression analysis are discussed. The following are input variables: ambient temperature, working pressure, steam production, and the type of fuel used in around 20 industrial boilers. Each boiler's emission data was collected using a TESTO 350 Combustion Gas Analyzer. The modeling, with a machine learning approach using the Gradient Boosting Regression algorithm, showed better performance in the predictions made on the test data, outperforming all other models studied. It was achieved with predicted values showing a mean absolute error of 0.51 and a coefficient of determination of 99.80%. Different regression models (DNN, MLR, RFR, GBR) were compared to select the most optimal. Compared to models based on Linear Regression, the DNN model has better prediction performance. The proposed model provides a new method to predict CO2, CO, NOx emissions, and exhaust gas outlet temperature.

Composite Fins Subsonic Flutter Prediction Based on Machine Learning

In the present work, the potential application of machine learning techniques in the flutter prediction of composite materials missile fins is investigated. The flutter velocity data set required for different fin aerodynamic geometries and materials is generated using a hybrid data collection method: from the wind tunnel experiments at flows ranging from 5 to 30 m/s at Re = 300,000 to 500,000, whereas synthetic data is collected using modified NACA flutter boundary model. Once the flutter data are collected, different regression algorithms were investigated, and the results were compared in terms of accuracy (when compared to the experimentally obtained results and the numerical flutter models), training time (minimization), and R-squared values (maximization). The algorithms investigated and their performance analyzed are fast forest regression, SDCA regression (stochastic dual coordinate ascent), and the light GBM (light gradient boosting machine) regression algorithm that belongs to the gradient boosting regression algorithm family. It was found that the light GBM algorithm renders the most accurate results. Based on this research, it can be concluded that artificial intelligence (machine learning) techniques can be successfully deployed in the analysis of complex flutter phenomena.

A novel approach to estimate the weight of food items based on features extracted from an image using boosting algorithms

Managing daily nutrition is a prominent concern among individuals in contemporary society. The advancement of dietary assessment systems and applications utilizing images has facilitated the effective management of individuals' nutritional information and dietary habits over time. The determination of food weight or volume is a vital part in these systems for assessing food quantities and nutritional information. This study presents a novel methodology for evaluating the weight of food by utilizing extracted features from images and training them through advanced boosting regression algorithms. Α unique dataset of 23,052 annotated food images of Mediterranean cuisine, including 226 different dishes with a reference object placed next to the dish, was used to train the proposed pipeline. Then, using extracted features from the annotated images, such as food area, reference object area, food id, food category, and food weight, we built a dataframe with 24,996 records. Finally, we trained the weight estimation model by applying cross validation, hyperparameter tuning, and boosting regression algorithms such as XGBoost, CatBoost, and LightGBM. Between the predicted and actual weight values for each food in the proposed dataset, the proposed model achieves a mean weight absolute error 3.93 g, a mean absolute percentage error 3.73% and a root mean square error 6.05 g for the 226 food items of the Mediterranean Greek Food database (MedGRFood), setting new perspectives in food image-based weight and nutrition estimate models and systems.

Accelerated screening of sensitive and selective MoO3-based gas sensing materials by combining first-principles and machine learning approach

A large amount of hybrid metal oxides can be proposed for detecting various gases from rapidly developing of 2D materials and elemental doping technologies, and the key issue is how to search the large parameter space in an efficient way. First-principles approach can calculate molecular-scale electronic properties and provide a guide of material design for experiments, however, it is too time-consuming to explore all the possible metal oxides doped with different elements. Herein, we develop a novel framework via combining first-principles and machine learning, with the aim of enabling more efficient and practical screening of MoO3-based gas sensors than by experiments or first-principles alone. Owing to the high accuracy of first-principles calculations verified by experimental results, we demonstrate the reliability of the proposed methods for evaluating gas sensing performance. By proposing a set of new descriptors including d-band center and average bond length with demanding low-cost calculations, we significantly reduce the amount of required training data. In particular, the gradient boosting regression algorithm exhibits highR-square value of 0.96 and low mean absolute error value of 0.22, indicating superior reliability of the model. This work opens an avenue for quickly screening of novel metal oxides and other nano-materials with superior sensitivity and selectivity for gas detection.

Machine Learning Design of Single-Atom Catalysts for Nitrogen Fixation.

First-principles calculations have been combined with machine learning in the design of transition-metal single-atom catalysts. Readily available descriptors are selected to describe the nitrogen activation capability of metals and coordinating atoms. Thus, a series of V/Nb/Ta-Nx single-atom catalysts are screened out as promising structures upon considering the stability, activity, and selectivity investigated computationally. Furthermore, by using the gradient boosting regression algorithm, an accurate prediction of the hydrogenation barriers for the nitrogen reduction reaction (NRR) is achievable, with a root-mean-squared error of 0.07 eV. The integration of high-throughput computation and machine learning constitutes a powerful strategy for the acceleration of catalyst design. This approach facilitates the rapid and accurate prediction of the NRR performance of more than 1000 single-atom catalyst structures. Moreover, the current work provides further insights by elaborately correlating the structure and performance, which may be instructive for both the design and application of vanadium-group catalysts.

Estimation of Species-Scale Canopy Chlorophyll Content in Mangroves from UAV and GF-6 Data

The growth of mangroves is inhibited due to environmental degradation, and changes in the growing health of mangrove forests cause changes in internal physicochemical parameters. The canopy chlorophyll content is an important indicator to monitor the health status of mangroves. We study the effective inversion data sources and methods of mangrove health indicator parameters to monitor the health of mangrove ecosystems in typical areas of Beibu Gulf, Guangxi. In this study, we evaluated the capability of UAV, GF-6 data, and machine learning regression algorithms in estimating mangrove species-scale canopy chlorophyll content (CCC). Effective measures for mangrove pest and disease pressure, Sporobolus alterniflorus invasion, and anthropogenic risk are also explored, which are important for mangrove conservation and restoration. (1) We obtained several feature variables by constructing a combined vegetation index, and the most sensitive band of mangrove CCC was selected by the characteristic variable evaluation, and the CCC model at the mangrove species-scale was evaluated and validated. Through variable preferences, the feature variables with the highest contribution of Avicennia marina, Aegiceras corniculatum, Kandelia candel, and a collection of three categories of species in the UAV data were indices of RI35, MDATT413, RI35, and NDI35. (2) Random Forest, Gradient Boosting Regression Tree, and Extreme Gradient Boosting were evaluated using the root-mean-square error and coefficient of determination accuracy metrics. Extreme Gradient Boosting regression algorithms were evaluated for accuracy. In both UAV data and GF-6, RF achieved optimal results in inverse mangrove Aegiceras corniculatum species CCC, with higher stability and robustness in machine learning regressors. (3) Due to the sparse distribution of Kandelia candel in the study area and the low spatial resolution of the images, there is an increased possibility that individual image elements contain environmental noise, such as soil. Therefore, the level of CCC can effectively reflect the health status of mangroves and further reflect the increased possibility of the study area being exposed to risks, such as degradation. The establishment of the current protected areas and restoration of degraded ecosystems are effective measures to cope with the risks of mangrove pest and disease stress, invasion of Sporobolus alterniflorus, and anthropogenic activities, which are important for the protection and restoration of mangroves. This study provides an important data reference and risk warning for mangrove restoration and conservation.

Machine learning-aided prediction of nitrogen heterocycles in bio-oil from the pyrolysis of biomass

Nitrogen heterocyclic compounds in bio-oil (NH_Oil) made from biomass pyrolysis such as pyrroles, pyrazines, and indoles, have a relative content of 0–30%. NH_Oil is a NOx precursor if bio-oil is used as a fuel, but it has a high potential as a precursor for high-value chemicals. However, predicting and controlling NH_Oil are challenging because of the complexity of the pyrolysis reaction system. Machine learning (ML) shows significant potential for addressing this issue. In this study, the relative contents of NH_Oil, 5-membered NH_Oil, 6-membered NH_Oil, bio-oil yield, and the content of nitrogen in bio-oil were predicted using Random Forest and gradient boosting regression algorithms, with test regression coefficients values of 0.77–0.87 and 0.74–0.81 obtained for the former and latter ML models, respectively. Biomass N was the most important factor in predicting the bio-oil yield, whereas biomass N/C was the most significant of the other four targets and can be used as a proxy to assess the potential of biomass feedstock as fuel material or N-containing value-added chemical precursor. The optimization of pyrolysis parameters within ML models provides useful information for instructing experimental studies, indicating ML-aided bio-oil prediction and engineering show great promise and are worthy of further investigation.

Combined machine learning forecasting method for short-term power load based on the dynamic weight adjustment

The load of power system exhibits evident characteristics of volatility and randomness. The traditional load forecasting algorithm usually studies and trains the historical data to obtain the load model, which makes it difficult to adapt to the load dynamic change situation, and then resulting the unreasonable inaccurate prediction. In this paper, a combinatorial machine learning model is adopted to forecast short-term power load using a dynamic adjustable weight. Firstly, a combined machine learning model is constructed using three types of algorithms including the improved long and short-term neural network, bagging algorithm, and boosting regression algorithm. The weight of each algorithm is determined dynamically by the improved error function. Secondly, the dynamic error function and the optimal weight optimization algorithm are employed so as to balance the contradiction between the speed and accuracy of dynamic adjustment. For different months or different days within a month, different weight adjustment algorithms are selected for enhancement. In addition, a penalty term is introduced to improve the algorithm accuracy and the final prediction outcomes. Finally, a practical load prediction case is simulated and compared with the traditional combined prediction model with fixed weights. It is verified that the proposed model can effectively eliminate the excessive errors caused by the poor dynamic response effect. It has a good dynamic response effect and accurate prediction. The error rate is only 1.24% when the load fluctuation is significant. This study provides a novel approach to forecasting short-term power load.

Deep learning method for rapidly estimating pig body size

Context During pig breeding, a change in a pig’s body size is an important indicator that reflects its health. However, it is difficult to extract the necessary features from images to estimate pig body size without contact. Aims It is crucial to develop a fast and accurate body size estimation algorithm to meet the practical needs of farms, i.e., numerous body size detections. Methods This report presents a rapid pig body size estimation technique based on deep learning. The YOLOv5 model is enhanced by integrating MobilenetV3, and a lightweight object detection network is introduced as the feature extraction network. An attention mechanism is also added to this system. Following these improvements, the proposed YOLOv5_Mobilenet_SE model is more suitable for the small-target detection of key parts of live pigs. A depth camera was used at a fixed height to capture the pig’s back information, which enables calculations of the critical height, i.e., the body height, of live pigs. Other key measuring points on the pig are generated according to the detection frame of the key parts located by the model. A gradient boosting regression algorithm is used to establish the body size prediction model based on the Euclidean distance between the key measuring points and the actual body size data. Key results The upgraded YOLOv5_Mobilenet_SE model achieves a mean average precision of 3.9%, which is higher than that obtained using the original YOLOv5 model. The model size is reduced from 91.2 to 10.2 M, and the average detection time for each image is 4.4 ms. The mean absolute percent errors in terms of body size, body width, and body height are 2.02%, 1.95%, and 1.84%, respectively, relative to manual measurements. Conclusions This method greatly reduces the model size and detection time while ensuring accuracy, and therefore, this method can cut costs for farms performing pig body size measurements. Implications The results of this study can provide technical support for automated and digital monitoring in the pig breeding industry.

Solar Irradiance Probabilistic Forecasting Using Machine Learning, Metaheuristic Models and Numerical Weather Predictions

Solar-power-generation forecasting tools are essential for microgrid stability, operation, and planning. The prediction of solar irradiance (SI) usually relies on the time series of SI and other meteorological data. In this study, the considered microgrid was a combined cold- and power-generation system, located in Tahiti. Point forecasts were obtained using a particle swarm optimization (PSO) algorithm combined with three stand-alone models: XGboost (PSO-XGboost), the long short-term memory neural network (PSO-LSTM), and the gradient boosting regression algorithm (PSO-GBRT). The implemented daily SI forecasts relied on an hourly time-step. The input data were composed of outputs from the numerical forecasting model AROME (Météo France) combined with historical meteorological data. Our three hybrid models were compared with other stand-alone models, namely, artificial neural network (ANN), convolutional neural network (CNN), random forest (RF), LSTM, GBRT, and XGboost. The probabilistic forecasts were obtained by mapping the quantiles of the hourly residuals, which enabled the computation of 38%, 68%, 95%, and 99% prediction intervals (PIs). The experimental results showed that PSO-LSTM had the best accuracy for day-ahead solar irradiance forecasting compared with the other benchmark models, through overall deterministic and probabilistic metrics.

Machine learning model to predict endophytic colonisation of rice cultivar plant tissues by Beauveria bassiana isolates and their potential as bio-control agents against rice stem borer using existing knowledge

Background: Finding well-known Beauveria bassiana isolates that could preserve rice crops from Sesamia calamistis (stem borer) is problematic. Another difficult task is the development of precise inoculation methods, which have been employed for their establishment as endophytes in cereal crops. This study proposed machine learning models to predict the best entomopathogenic fungi, Beauveria bassiana that could directly protect rice crops against Sesamia calamistis. Methods: Data driven machine learning decisions were implemented and assessed from 60 experimental runs with nine different feature/input variables and three target/output variables following foliar spray and seed treatment inoculation method. The feature variables consisted of rice plant tissue, such as Nerica-L19, Nerica1, Nerica8, the time, and the five promising isolates Beauveria bassiana (Bb3, Bb4, Bb10, Bb21, Bb35). The target variable consisted of the number of colonised roots, stems and leaves, expressed as a percentage depending on the degree of protection after each inoculation. A data driven decision by the extreme gradient boosting regression algorithm was used to proficiently abstract the situation where there is no direct relationship between features and target variables. Results: The foliar spray inoculation method exhibited high coefficient of determination (R2) of 0.99, 0.98 and 0.94 depending on the number of colonised stems, roots and leaves, respectively, while the seed treatment approach exhibited the coefficient of determination (R2) of 0.91, 0.87 and 0.75, respectively. Conclusions: These results demonstrated that the Extreme Gradient Boosting algorithm effectively abstracted the nonlinear relationship between the attribute variables that were taken into consideration and predicted Beauveria bassiana as a bio-pesticide for rice and perhaps other cereal stem borers. Thus, this XGBoost regression model could be used to navigate the optimization domain and reduce the development time of the biocontrol process.

Satellite data-driven and knowledge-informed machine learning model for estimating global internal solitary wave speed

Internal solitary waves (ISW) are widely distributed worldwide and significantly affect the ocean environment and offshore activities. ISW propagation speed is important for ISW forecasts and varies largely globally. This study collected 810 quasi-synchronous optical satellite images with clear ISW signatures in 13 global hotspots to build a large ISW dataset. ISW speed was calculated using extracted ISW wave crest locations and the time difference between satellite image pairs. The dataset contains 57,196 samples, including extracted ISW wave crests and corresponding ISW phase speed. We developed an ISW propagation speed (IPS) model based on the dataset using machine learning techniques. The model structure includes clustering and regression algorithms. The model adopts two tailored modifications to incorporate the ISW domain knowledge and solve the ISW sample distribution imbalance problems. Implementation domain knowledge (IDK) includes selecting relevant ocean factors and ISW properties based on oceanography theory and remote sensing imaging mechanisms. The second tailored modification is adopting advanced model architecture (AMA) by introducing the Gaussian clustering algorithm to classify ISW samples into several groups beyond the limitation of space and time. The extreme gradient boosting regression algorithm was applied in each group to build the IPS model. We used 47,425 samples as the training dataset and the remaining 9771 samples as the test dataset. The model-predicted ISW speed shows good accuracy, with a root mean square error/relative error rate (RER) of 0.16 (7.9) and 0.30 m/s (12.7%) on the training and test dataset. Analysis shows that IDK and AMA improve the model performance by 19.4% and 13.1%, respectively. With a one-pixel error in the peak-to-peak distance of input parameters, the model results degraded from 0.30 m/s to 0.33 m/s. The IPS model was applied to estimate ISW speeds in ocean regions besides the 13 hotspots, and the average RER is 6.0%. ISW forecast in seven ocean areas was tested, and the results indicate that the IPS model can describe ISW propagation patterns. The model results reveal that the ISW phase speed strongly correlates with the spring and neap tide. The IPS model results show that ISW speed is decreased with a deepening stratification. Model-predicted global ISW propagation speed comparison shows that the Celebes Sea and North-West of South America has the fastest and slowest propagating ISWs all year around, respectively. Discussion on the background current's influence on the IPS model results is presented.

High-quality fracture network mapping using high frequency logging while drilling (LWD) data: MSEEL case study

The Marcellus Shale and Energy Environmental Laboratory (MSEEL) provides a comprehensive dataset and field tests that can be used to study the significance of preexisting natural fractures in different subsurface engineering problems such as the effectiveness of the stimulation of an unconventional reservoir, optimized geothermal fluids movement and integrity of the CO2 storage site. Conventionally natural fracture intensity is obtained using sonic and micro-resistivity imaging logs. However, these techniques significantly suffer from two major deficiencies: Human bias in log interpretation and extremely long interpretation time. These two deficiencies are well-recognized in the industry; however, no standard procedures exist to address them. In this study, a new automated machine learning workflow (AMLW) is introduced that uses the LWD high-resolution acceleration data along the horizontal laterals to predict the natural fracture intensities originally obtained using sonic and micro-resistivity imaging. The accuracy and robustness of the new workflow to predict the near wellbore fracture intensities are tested using both regression and classification approaches. Both the regression and classification approaches were able to predict the fracture intensities with high accuracy (average Mean Squared Error of 0.0085 for regression and average accuracy of 0.94 in the confusion matrix for classification). We have shown that only 10%–15% of the labeled resistivity image log is required for training and validation of the machine-learning model. The Automated workflow resulted in K-Neighbors Regressor and classifier algorithms as the best algorithms with a 52.74% and 139.3% improvement in comparison to the Gradient Boosting Regression algorithm (i.e. the fifth best algorithm).

Dynamic monitoring of phycocyanin concentration in Chaohu Lake of China using Sentinel-3 images and its indication of cyanobacterial blooms

• Gradient boosting regression has the great potential in water quality retrieval. • PC concentration in Chaohu Lake has a high temporal consistency with CBs proportion. • PC concentration has the indicative significance for early warning of CBs outbreak. Phycocyanin (PC) is an indicator pigment of cyanobacteria in water. Monitoring the dynamic variation of PC concentration in lakes by satellite remote sensing is of great significance for the effective prevention and control of cyanobacterial blooms (CBs). Few studies have quantitatively monitored the long-term PC concentration variation in water and explored the relationship between PC concentration and CBs based on the advantages of the high temporal and spectral resolution of Sentinel-3 OLCI images. In this study, a retrieval model of PC concentration suitable for Sentinel-3 OLCI images was constructed and applied to the Sentinel-3 image dataset from May 2016 to November 2021 to analyze the spatiotemporal PC concentration variations in Chaohu Lake of China. The factors influencing PC concentration and its indicative significance for CBs outbreak were also explored. The results show that the PC concentration retrieval model based on the gradient boosting regression algorithm has the best performance (R 2 = 0.86, root mean square error = 45.44 ug/L, and mean absolute percentage error = 26.27 %), which has great application potential in the retrieval of other water quality parameters. In summer and autumn, the PC concentration in western Chaohu Lake was higher than that in other lake parts, which was mainly related to the higher nutrient concentration and prevailing wind direction in this part. During 2016–2021, the average PC concentration in Chaohu Lake in 2019 was the highest, and that in 2017 was the lowest, mainly affected by air temperature and precipitation. The variation of PC concentration in Chaohu Lake has a high temporal consistency with the proportion of CBs area, and the PC concentration of 303.9 μg/L is a necessary condition for CBs outbreak. This study revealed that Sentinel-3 OLCI images could provide a useful data source for high-frequency dynamic monitoring of PC concentrations in lakes and reservoirs, and the quantitative monitoring of PC concentrations in water has relevant indicative significance for early warning of CBs in inland water.

Can data reliability of low-cost sensor devices for indoor air particulate matter monitoring be improved? – An approach using machine learning

Poor indoor air quality has adverse health impacts. Children are considered a risk group, and they spend a significant time indoors at home and in schools. Air quality monitoring has traditionally been limited due to the cost and size of the monitoring stations. Recent advancements in low-cost sensors technology allow for economical, scalable and real-time monitoring, which is especially helpful in monitoring air quality in indoor environments, as they are prone to sudden peaks in pollutant concentrations. However, data reliability is still a considerable challenge to overcome in low-cost sensors technology. Thus, following a monitoring campaign in a nursery and primary school in Porto urban area, the present study analyzed the performance of three commercially available low-cost IoT devices for indoor air quality monitoring in real-world against a research-grade device used as a reference and developed regression models to improve their reliability. This paper also presents the developed on-field calibration models via machine learning technique using multiple linear regression, support vector regression, and gradient boosting regression algorithms and focuses on particulate matter (PM1, PM2.5, PM10) data collected by the devices. The performance evaluation results showed poor detection of particulates in classrooms by the low-cost devices compared to the reference. The on-field calibration algorithms showed a considerable improvement in all three devices' accuracy (reaching up to R2 > 0.9) for the light scattering technology based particulate matter sensors. The results also show the different performance of low-cost devices in the lunchroom compared to the classrooms of the same school building, indicating the need for calibration in different microenvironments.

Diffuse Imaging Approach for Universal Noninvasive Blood Glucose Measurements

We proposed a diffuse imaging approach for universal noninvasive blood glucose measurements based on visible light, which can predict the blood glucose concentration without personal calibration. The proposed approach used a CCD to obtain diffuse images from human index finger pulp. The denoising autoencoder algorithm adopted effectively extracted the scattering information highly related to blood glucose concentration from the diffuse images, and the gradient boosting regression algorithm enabled an accurate calculation of blood glucose concentration without prior personalized calibration.In vivoexperimental results showed that the proposed approach had a mean absolute error of 19.44 mg/dl, with all the predicted results observed within the clinically acceptable region (Region A: 78.9%) in the Clarke error grid analysis. Compared to other blood glucose concentration measurement methods of scattering coefficient, this new method does not require individual calibration, therefore it is easier to implement and popularize, which is critical for the noninvasive monitoring of blood glucose concentration.