Locally Weighted Learning Research Articles

Prediction of air compressor faults with feature fusion and machine learning

Air compressors are critical for many industries, but early detection of faults is crucial for keeping them running smoothly and minimizing maintenance costs. This contribution investigates the use of predictive machine learning models and feature fusion to diagnose faults in single-acting, single-stage reciprocating air compressors. Vibration signals acquired under healthy and different faulty conditions (inlet valve fluttering, outlet valve fluttering, inlet-outlet valve fluttering, and check valve fault) serve as the study’s input data. Diverse features including statistical attributes, histogram data, and auto-regressive moving average (ARMA) coefficients are extracted from the vibration signals. To identify the most relevant features, the J48 decision tree algorithm is employed. Five lazy classifiers viz. k-nearest neighbor (kNN), K-star, local kNN, locally weighted learning (LWL), and random subspace ensemble K-nearest neighbors (RseslibKnn) are then used for fault classification, each applied to the individual feature sets. The classifiers achieve commendable accuracy, ranging from 85.33% (K-star and local kNN) to 96.00% (RseslibKnn) for individual features. However, the true innovation lies in feature fusion. Combining the three feature types, statistical, histogram, and ARMA, significantly improves accuracy. When local kNN is used with fused features, the model achieves a remarkable 100% classification accuracy, demonstrating the effectiveness of this approach for reliable fault diagnosis in air compressors.

Prediction of white blood cell count during exercise: a comparison between standalone and hybrid intelligent algorithms

Decades of research in exercise immunology have demonstrated the profound impact of exercise on the immune response, influencing an individual's disease susceptibility. Accurate prediction of white blood cells (WBCs) count during exercise can help to design effective training programs to maintain optimal the immune system function and prevent its suppression. In this regard, this study aimed to develop an easy-to-use and efficient modelling tool for predicting WBCs count during exercise. To achieve this goal, the predictive power of a range of machine-learning algorithms, including six standalone models (M5 prime (M5P), random forest (RF), alternating model trees (AMT), reduced error pruning tree (REPT), locally weighted learning (LWL), and support vector regression (SVR)) were assessed along with six types of hybrid models trained with a bagging (BA) algorithm (BA-M5P, BA-RF, BA-AMT, BA-REPT, BA-LWL, and BA- SVR). A comprehensive database was constructed from 200 eligible people. The models employed post-exercise training WBCs counts as the output parameter and seven WBCs-influencing factors, including intensity and duration of exercise, pre-exercise training WBCs counts, age, body fat percentage, maximal aerobic capacity, and muscle mass as input parameters. Comparing the prediction results of the models to the observed WBCs using standard statistics indicated that the BA-M5P model had the greatest potential to produce a robust prediction of the number of lymphocytes, neutrophils, monocytes, and WBC compared to other models. Moreover, pre-exercise training WBCs counts, intensity and duration of exercise and body fat percentage were the most important features in predicting WBCs counts. These findings hold significant implications for the advancement of exercise immunology and the promotion of public health.

Using machine learning algorithms for predicting bridge vertical deflection

Machine learning (ML) is being increasingly used to ease structural health monitoring and management of bridges. Therefore, the main objective of this study is to apply two state-of-the-art ML algorithms, namely locally weighted learning (LWL) and K-Star for the prediction of vertical deflection of composite bridges. To accomplish the objective, 83 track loading tests were carried out at various bridges located in Vietnam and deflection data was collected for models’ development. Model's validation and comparison were carried out using different popular methods, namely mean absolute error, root mean squared error and R on both training (70%) and validation (30%) datasets. The results of this study indicate that the K-Star algorithm outperforms LWL in predicting the vertical deflection of composite bridges. Consequently, K-Star can serve as an effective tool for rapidly predicting vertical deflection, thereby facilitating more efficient bridge health monitoring and management. This efficiency can lead to significant time and cost savings in maintaining the structural integrity of composite bridges.

Development of web-based learning application to enhance understanding in Philippine history

This study introduces a web-based learning system designed to enhance the recall and understanding of Philippine history through engaging quizzes and games, aligning with ISO 25010 standards. The system prioritizes functional suitability, usability, reliability, performance efficiency, and portability, with the aim of aiding users, particularly students, in remembering and studying Philippine history. Leveraging interactive features such as game content and quizzes, the system actively engages learners, promoting comprehension and retention of historical facts. Instant feedback mechanisms further contribute to users' ability to identify strengths and weaknesses in understanding Philippine history. To evaluate the system's performance, the researchers applied the Locally Weighted Learning (LWL) algorithm, adhering to ISO 25010 guidelines. The LWL algorithm, combining lazy learning and a decision stump classifier with linear weighting kernels, achieved an impressive 90% accuracy on the "Book1" dataset. This comprehensive evaluation validated the system's reliability and effectiveness, showcasing high accuracy in assessing attributes such as data accuracy, system availability, user-friendliness, responsiveness, and security assurance. The study also includes a regression analysis, indicating a moderate positive linear relationship (multiple R of 0.4243) between independent variables (functional suitability, usability, performance efficiency, reliability, and portability) and the dependent variable (software quality score). While the model explains approximately 22.38% of the variability, the non-significant p-value in the ANOVA table prompts consideration for further refinement. Overall, this research contributes to the field of educational technology by providing valuable insights into the development and evaluation of a history-focused learning application, emphasizing both practical usability and theoretical reliability within the framework of ISO 25010.

Landslide susceptibility assessment using locally weighted learning integrated with machine learning algorithms

Assessing landslide susceptibility and predicting the possibility of landslide event is the foundation and prerequisite for emergency response and management of landslide disaster. The target of current paper is to propose five integration models based on integrating locally weighted learning (LWL) with a radial basis function classifier (RBF), Fisher's linear discriminant (FLDA), quadratic discriminant analysis (QDA), a Credal decision tree (CDT), an alternating decision tree (ADT) and random subspace (RS) and the performance of five integration models were compared for modeling landslide susceptibility. Yongxin County from China was employed as a study area, 364 landslide locations and fifteen environmental factors were applied. The results demonstrate that the proposed LWL-RS-ADT model is more reliable and stable than the other models. Among the fifteen environmental factors, NDVI, lithology, and altitude are the very significant factors in the six models. It is concluded that the proposed integration models provide an effective way to predict the susceptibility of landslides.

Machine Learning Prediction of Time Series Covid-19 Data in West Java, Indonesia

In 2019, the COVID-19 pandemic appeared. There have been several efforts to curb the spread of this virus. West Java, Indonesia, employs social restrictions to prevent the spread of this disease. However, this method destroyed the economy of the people. If no instances were detected in the region, the World Health Organization (WHO) authorized the social restrictions to be relaxed. If the government lifts the social limitation, the decision must also consider the potential of future confirmed instances. By utilizing machine learning, it is possible to forecast future data. This work utilized the following algorithms: linear regression (LR), locally weighted learning (LWL), multi-layer perceptron (MLP), radial basis function regression (RBF), and support vector machine (SVM). The study investigated daily new instances of COVID-19 in West Java, Indonesia, from March 2, 2020, to October 15, 2020. The RBF algorithm was the best in this investigation. Mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and relative absolute error (RAE) were 48.85, 89.73, 88.67, 62.99, and 60.88, respectively. The RBF prediction model may be proposed to the government of West Java for assessing data on COVID-19 instances, particularly in social restriction management. It is anticipated that West Java would have a minimum of 275 new cases every day for the following 30 days beginning on October 16, 2020. Consequently, the easing of societal limitations requires careful consideration.

Developing a smart and clean technology for bioremediation of antibiotic contamination in arable lands

This study presents a smart technological framework to efficiently remove azithromycin from natural soil resources using bioremediation techniques. The framework consists of several modules, each with different models such as Penicillium Simplicissimum (PS) bioactivity, soft computing models, statistical optimisation, Machine Learning (ML) algorithms, and Decision Tree (DT) control system based on Removal Percentage (RP). The first module involves designing experiments using a literature review and the Taguchi Orthogonal design method for cultural conditions. The RP is predicted as a function of cultural parameters using Response Surface Methodology (RSM) and three ML algorithms: Instance-Based K (IBK), KStar, and Locally Weighted Learning (LWL). The sensitivity analysis shows that pH is the most important factor among all parameters, including pH, Aeration Intensity (AI), Temperature, Microbial/Food (M/F) ratio, and Retention Time (RT), with a p-value of <0.0001. AI is the next most significant parameter, also with a p-value of <0.0001. The optimal biological conditions for removing azithromycin from soil resources are a temperature of 32 °C, pH of 5.5, M/F ratio of 1.59 mg/g, and AI of 8.59 m3/h. During the 100-day bioremediation process, RP was found to be an insignificant factor for more than 25 days, which simplifies the conditions. Among the ML algorithms, the IBK model provided the most accurate prediction of RT, with a correlation coefficient of over 95%.

Locally Weighted Learning Robot Control With Improved Parameter Convergence

An adaptive robot control approach that can achieve an accurate approximation of the complex robot dynamics without knowledge unlearning in the case of multiple sequential tasks is proposed in this article. A locally weighted learning (LWL) model with automatic structure growth according to the size of the learning domain is introduced to approximate unknown robot dynamics, and a composite learning technique with regressor extension is applied to improve parameter convergence. The LWL ensures that learning in one area of the learning domain does not cause unlearning in another area, and the composite learning theoretically guarantees the identification performance of the LWL model. By the exploitation of both stored data in memory and instantaneous data, parameter convergence in the LWL model is achieved under a more achievable interval-excitation condition than the stringent persistent-excitation condition. This further enhances the trajectory tracking performance for multiple sequential tasks, which is generally not achievable by existing approximation-based adaptive robot control approaches. Experimental studies have verified the superiority of the proposed approach over prevalent approaches.

The Use of Fluorescence Spectroscopic Data and Machine-Learning Algorithms to Discriminate Red Onion Cultivar and Breeding Line

The objective of this study was to evaluate differences between the red onion cultivar and breeding line using models based on selected fluorescence spectroscopic data built using machine-learning algorithms from different groups of Trees, Functions, Bayes, Meta, Rules, and Lazy. The combination of fluorescence spectroscopy and machine learning is an original approach to the non-destructive and objective discrimination of red onion samples. The selected fluorescence spectroscopic data were used to build models using algorithms from the groups of Trees, Functions, Bayes, Meta, Rules, and Lazy. The most satisfactory results were obtained using J48 and LMT (Logistic Model Tree) from the group of Trees, Multilayer Perceptron, and QDA (Quadratic Discriminant Analysis) from Functions, Naive Bayes from Bayes, Logit Boost from Meta, JRip from Rules, and LWL (Locally Weighted Learning) from Lazy. The average accuracy of discrimination of onion bulbs belonging to ‘Asenovgradska kaba’ and a red breeding line equal to 100% was found in the case of models developed using the LMT, Multilayer Perceptron, Naive Bayes, Logit Boost, and LWL algorithms. The TPR (True Positive Rate), Precision, and F-Measure of 1.000 and FPR (False Positive Rate) of 0.000, as well as the Kappa statistic of 1.0, were determined. The results revealed the usefulness of the approach combining fluorescence spectroscopy and machine learning to distinguish red onion cultivars and breeding lines.

Study of micro-porosity in electron beam butt welding

As complete elimination of porosity from the weld is very difficult, the next option available is to minimize this weld porosity, which is crucial for the safe performance of the welded components. However, this investigation through experiments alone is very tedious and time consuming. Additionally, very limited models are available in the literature for accurate prediction of different porosity attributes. The present study, thus, addressees both the experimental as well as modelling aspect on the study of micro-porosity during electron beam welding (EBW) of SS304 plates. Welding parameters are reported to have significant influence on the micro-porosity. Hence, the influences of these parameters on micro-porosity attributes, namely pores number, average diameter, and sphericity are extensively studied experimentally employing optical microscopy (OM), scanning electron microscopy (SEM), X-ray computed tomography (XCT), and Raman spectroscopy. This is followed by an elaborate modelling using seven popular and well-recognized machine learning algorithms (MLAs), namely multi-layer perceptron (MLP), support vector regression (SVR), M5P model trees regression, reduced error pruning tree (REPTree), random forest (RF), instance-based k-nearest neighbor algorithm (IBk), and locally weighted learning (LWL). These different techniques enhance the chance of obtaining the better predictions of the said micro-porosity attributes by overcoming the effect of data-dependence and other limitations of individual MLAs. The different model-predicted micro-porosity data are also validated through experimental data. Statistical tests and Monte-Carlo reliability analysis are additionally utilized to evaluate the performances of the employed algorithms. IBk and MLP are overall found to perform well.

Multi-step ahead soil temperature forecasting at different depths based on meteorological data: Integrating resampling algorithms and machine learning models

Direct soil temperature (ST) measurement is time-consuming and costly; thus, the use of simple and cost-effective machine learning (ML) tools is helpful. In this study, ML approaches, including KStar, instance-based K-nearest learning (IBK), and locally weighted learning (LWL), coupled with resampling algorithms of bagging (BA) and dagging (DA) (BA-IBK, BA-KStar, BA-LWL, DA-IBK, DA-KStar, and DA-LWL) were developed and tested for multi-step ahead (3, 6, and 9 d ahead) ST forecasting. In addition, a linear regression (LR) model was used as a benchmark to evaluate the results. A dataset was established, with daily ST time-series at 5 and 50 cm soil depths in a farmland as models’ output and meteorological data as models’ input, including mean (Tmean), minimum (Tmin), and maximum (Tmax) air temperatures, evaporation (Eva), sunshine hours (SSH), and solar radiation (SR), which were collected at Isfahan Synoptic Station (Iran) for 13 years (1992–2005). Six different input combination scenarios were selected based on Pearson's correlation coefficients between inputs and outputs and fed into the models. We used 70% of the data to train the models, with the remaining 30% used for model evaluation via multiple visual and quantitative metrics. Our findings showed that Tmean was the most effective input variable for ST forecasting in most of the developed models, while in some cases the combinations of variables, including Tmean and Tmax and Tmean, Tmax, Tmin, Eva, and SSH proved to be the best input combinations. Among the evaluated models, BA-KStar showed greater compatibility, while in most cases, BA-IBK and -LWL provided more accurate results, depending on soil depth. For the 5 cm soil depth, BA-KStar had superior performance (i.e., Nash-Sutcliffe efficiency (NSE) = 0.90, 0.87, and 0.85 for 3, 6, and 9 d ahead forecasting, respectively); for the 50 cm soil depth, DA-KStar outperformed the other models (i.e., NSE = 0.88, 0.89, and 0.89 for 3, 6, and 9 d ahead forecasting, respectively). The results confirmed that all hybrid models had higher prediction capabilities than the LR model.

Modelling and mapping of landslide susceptibility regulating potential ecosystem service loss: an experimental research in Saudi Arabia

The study aims to create a novel artificial intelligence model-based landslide susceptibility model (LSM) at Aqabat, Saudi Arabia. For LSM, a combination of bagging, dagging, random forest (RF) ensemble with locally weighted learning (LWL), viz. bagging-LWL, dagging-LWL, and RF-LWL has been developed. The 50 landslide areas were divided into two categories training (40) and testing (10). For training datasets, the LWL-Bagging model had the highest AUC value of ROC curve (AUC-0.91), followed by LWL-RF (AUC-0.881), LWL-Dagging (AUC-0.88) and LWL (AUC-0.875). For testing datasets, AUC values of ROC curve of 0.891, 0.876, 0.868, and 0.844 were found for LWL-Bagging, LWL-RF, LWL-Dagging, and LWL, respectively. Based on AUC value, all hybrid machines learning models performed well, but LWL-Bagging model performed as the best fit model. The potential ecosystem loss services were assessed by assessing possible landslide hazards. The research delivers reliable results to influence policy decisions on ecosystem management and landslide-assessment.

Clear-water scour depth prediction in long channel contractions: Application of new hybrid machine learning algorithms

Scour depth prediction and its prevention is one of the most important issues in channel and waterway design. However the potential for advanced machine learning (ML) algorithms to provide models of scour depth has yet to be explored. This study provides the first quantification of the predictive power of a range of standalone and hybrid machine learning models. Using previously collected scour depth data from laboratory flume experiments, the performance of five types of recently developed standalone machine learning techniques - the Isotonic Regression (ISOR), Sequential Minimal Optimization (SMO), Iterative Classifier Optimizer (ICO), Locally Weighted learning (LWL) and Least Median of Squares Regression (LMS) - are assessed, along with their hybrid versions with Dagging (DA) and Random Subspace (RS) algorithms. The main findings are five-fold. First, the DA-ICO model had the highest prediction power. Second, the hybrid models had a higher prediction power than standalone models. Third, all algorithms underestimated the maximum scour depth, except DA-ICO which predicted scour depth almost perfectly. Fourth, scour depth was most sensitive to densimetric particle Froude number followed by the non-dimensionalized contraction width, flow depth within the contraction, sediment geometric standard deviation, approach flow velocity and median grain size. Fifth, most of the algorithms performed best when all the input parameters were involved in the building of the model. An important exception was the best performing model that required only four input parameters: densimetric particle Froude number, non-dimensionalized contraction width, flow depth within the contraction and sediment geometric standard deviation. Overall the results revealed that hybrid machine learning algorithms provide more accurate predictions of scour depth than empirical equations and traditional ML-algorithms. In particular, the DA-ICO model not only created the most accurate predictions but also used the fewest easily and readily measured input parameters. Thus this type of model could be of real benefit to practicing engineers required to estimate maximum scour depth when designing in-channel structures.

Predicting stable gravel-bed river hydraulic geometry: A test of novel, advanced, hybrid data mining algorithms

Accurate prediction of stable alluvial hydraulic geometry, in which erosion and sedimentation are in equilibrium, is one of the most difficult but critical topics in the field of river engineering. Data mining algorithms have been gaining more attention in this field due to their high performance and flexibility. However, an understanding of the potential for these algorithms to provide fast, cheap, and accurate predictions of hydraulic geometry is lacking. This study provides the first quantification of this potential. Using at-a-station field data, predictions of flow depth, water-surface width and longitudinal water surface slope are made using three standalone data mining techniques -, Instance-based Learning (IBK), KStar, Locally Weighted Learning (LWL) - along with four types of novel hybrid algorithms in which the standalone models are trained with Vote, Attribute Selected Classifier (ASC), Regression by Discretization (RBD), and Cross-validation Parameter Selection (CVPS) algorithms (Vote-IBK, Vote-Kstar, Vote-LWL, ASC-IBK, ASC-Kstar, ASC-LWL, RBD-IBK, RBD-Kstar, RBD-LWL, CVPS-IBK, CVPS-Kstar, CVPS-LWL). Through a comparison of their predictive performance and a sensitivity analysis of the driving variables, the results reveal: (1) Shield stress was the most effective parameter in the prediction of all geometry dimensions; (2) hybrid models had a higher prediction power than standalone data mining models, empirical equations and traditional machine learning algorithms; (3) Vote-Kstar model had the highest performance in predicting depth and width, and ASC-Kstar in estimating slope, each providing very good prediction performance. Through these algorithms, the hydraulic geometry of any river can potentially be predicted accurately and with ease using just a few, readily available flow and channel parameters. Thus, the results reveal that these models have great potential for use in stable channel design in data poor catchments, especially in developing nations where technical modelling skills and understanding of the hydraulic and sediment processes occurring in the river system may be lacking.

Comparing of Data Mining Techniques for Predicting In-Hospital Mortality Among Patients with COVID-19

Introduction: The COVID-19 epidemic is currently fronting the worldwide health care systems with many qualms and unexpected challenges in medical decision-making and the effective sharing of medical resources. Machine Learning (ML)-based prediction models can be potentially advantageous to overcome these uncertainties. Objective: This study aims to train several ML algorithms to predict the COVID-19 in-hospital mortality and compare their performance to choose the best performing algorithm. Finally, the contributing factors scored using some feature selection methods. Material and Methods: Using a single-center registry, we studied the records of 1353 confirmed COVID-19 hospitalized patients from Ayatollah Taleghani hospital, Abadan city, Iran. We applied six feature scoring techniques and nine well-known ML algorithms. To evaluate the models’ performances, the metrics derived from the confusion matrix calculated. Results: The study participants were 1353 patients, the male sex found to be higher than the women (742 vs. 611), and the median age was 57.25 (interquartile 18-100). After feature scoring, out of 54 variables, absolute neutrophil/lymphocyte count and loss of taste and smell were found the top three predictors. On the other hand, platelet count, magnesium, and headache gained the lowest importance for predicting the COVID-19 mortality. Experimental results indicated that the Bayesian network algorithm with an accuracy of 89.31% and a sensitivity of 64.2 % has been more successful in predicting mortality. Conclusion: ML provides a reasonable level of accuracy in predicting. So, using the ML-based prediction models facilitate more responsive health systems and would be beneficial for timely identification of vulnerable patients to inform appropriate judgment by the health care providers. Abbreviation: Coronavirus Disease 2019 (COVID‐19), World Health Organization (WHO), Machine Learning (ML), Artificial Intelligence (AI), Multilayer Perceptron (MLP), Support Vector Machine (SVM), Locally Weighted Learning (LWL), Clinical Decision Support System (CDSS)

Locally Weighted Non-Parametric Modeling of Ship Maneuvering Motion Based on Sparse Gaussian Process

This paper explores a fast and efficient method for identifying and modeling ship maneuvering motion, and conducts a comprehensive experiment. Through the ship maneuvering test, the dynamics interaction between ship and the environment is obtained. Then, the LWL (Locally Weighted Learning algorithm) underlying architecture is constructed by sparse Gaussian Process to reduce the data requirements of LWL-based ship maneuvering motion modeling and to improve the performance for LWL. On this basis, a non-parametric model of ship maneuvering motion is established based on the locally weighted sparse Gaussian Process, and the traditional mathematical model of ship maneuvering motion is replaced by the generative model. This generative model considers the hydrodynamic effects of ships, and reduces the sensitivity of local weighted learning to sample data. In addition, matrix operations are transferred to the auxiliary platform to optimize the calculation performance of the method. Finally, the simulation results of ship maneuvering motion indicate that this method has the characteristics of efficiency, rapidity and universality, and its accuracy conforms to engineering practice.

Novel Ensemble Forecasting of Streamflow Using Locally Weighted Learning Algorithm

The development of advanced computational models for improving the accuracy of streamflow forecasting could save time and cost for sustainable water resource management. In this study, a locally weighted learning (LWL) algorithm is combined with the Additive Regression (AR), Bagging (BG), Dagging (DG), Random Subspace (RS), and Rotation Forest (RF) ensemble techniques for the streamflow forecasting in the Jhelum Catchment, Pakistan. To build the models, we grouped the initial parameters into four different scenarios (M1–M4) of input data with a five-fold cross-validation (I–V) approach. To evaluate the accuracy of the developed ensemble models, previous lagged values of streamflow were used as inputs whereas the cross-validation technique and periodicity input were used to examine prediction accuracy on the basis of root correlation coefficient (R), root mean squared error (RMSE), mean absolute error (MAE), relative absolute error (RAE), and root relative squared error (RRSE). The results showed that the incorporation of periodicity (i.e., MN) as an additional input variable considerably improved both the training performance and predictive performance of the models. A comparison between the results obtained from the input combinations III and IV revealed a significant performance improvement. The cross-validation revealed that the dataset M3 provided more accurate results compared to the other datasets. While all the ensemble models successfully outperformed the standalone LWL model, the ensemble LWL-AR model was identified as the best model. Our study demonstrated that the ensemble modeling approach is a robust and promising alternative to the single forecasting of streamflow that should be further investigated with different datasets from other regions around the world.

Stress classification based on human electromagnetic radiation analysis

&lt;p&gt;Stress is a feeling of emotional or physical tension due to any events that makes one feel frustrated, angry or nervous. It a situation that trigger particular biological response when encounter a threat or challenge. This paper discussed stress classification based on human electromagnetic radiation (EMR). At first, the collected radiation frequency data is analyzed using multivariate analysis of variance (MANOVA) to identify the significance points for the classification. Then, the data is classified using locally weighted learning (LWL) algorithm. The results show stress classification using EMR based on third eye and throat chakra points obtained accuracy of more than 60%.&lt;/p&gt;

Mapping forest fire susceptibility using spatially explicit ensemble models based on the locally weighted learning algorithm

Fire is among the most dangerous and devastating natural hazards in forest ecosystems around the world. The development of computational ensemble models for improving the predictive accuracy of forest fire susceptibilities could save time and cost in firefighting efforts. Here, we combined a locally weighted learning (LWL) algorithm with the Cascade Generalization (CG), Bagging, Decorate, and Dagging ensemble learning techniques for the prediction of forest fire susceptibility in the Pu Mat National Park, Nghe An Province, Vietnam. A geospatial database that contained records from 56 historical fires and nine explanatory variables was employed to train the standalone LWL model and its derived ensemble models. The models were validated for their goodness-of-fit and predictive capability using the area under the receiver operating characteristic curve (AUC) and several other statistical performance criteria. The CG-LWL and Bagging-LWL models with AUC = 0.993 showed the highest training performance, whereas the Dagging-LWL ensemble model with AUC = 0.983 performed better than Decorate-LWL (AUC = 0.976), CG-LWL and Bagging-LWL (AUC = 0.972), and LWL (AUC = 0.965) for predicting the spatial pattern of fire susceptibilities across the study area. Our study promotes the application of ensemble models in forest fire prediction and enhances the researchers' understanding of the processes of model building. Although these four ensemble models were originally developed for the estimation of forest fire susceptibility, the models are sufficiently general to be used for predicting other types of natural hazards, such as landslides, floods, and dust storms, by considering local geo-environmental factors.

Locally weighted learning based hybrid intelligence models for groundwater potential mapping and modeling: A case study at Gia Lai province, Vietnam

The groundwater potential map is an important tool for a sustainable water management and land use planning, particularly for agricultural countries like Vietnam. In this article, we proposed new machine learning ensemble techniques namely AdaBoost ensemble (ABLWL), Bagging ensemble (BLWL), Multi Boost ensemble (MBLWL), Rotation Forest ensemble (RFLWL) with Locally Weighted Learning (LWL) algorithm as a base classifier to build the groundwater potential map of Gia Lai province in Vietnam. For this study, eleven conditioning factors (aspect, altitude, curvature, slope, Stream Transport Index (STI), Topographic Wetness Index (TWI), soil, geology, river density, rainfall, land-use) and 134 wells yield data was used to create training (70%) and testing (30%) datasets for the development and validation of the models. Several statistical indices were used namely Positive Predictive Value (PPV), Negative Predictive Value (NPV), Sensitivity (SST), Specificity (SPF), Accuracy (ACC), Kappa, and Receiver Operating Characteristics (ROC) curve to validate and compare performance of models. Results show that performance of all the models is good to very good (AUC: 0.75 to 0.829) but the ABLWL model with AUC = 0.89 is the best. All the models applied in this study can support decision-makers to streamline the management of the groundwater and to develop economy not only of specific territories but also in other regions across the world with minor changes of the input parameters.