Wyniki wyszukiwania - BazTech

1

Classical and Quantum SVM for Electromyography-Based Myopathy Detection: A Comparative Exploration

Hammachi Radhouane, Messaoudi Noureddine, Belkacem Samia, Pasetto Edoardo, Delilbasic Amer

Polish Journal of Medical Physics and Engineering

|

2025

|

Vol. 31, Iss. 2

118--130

EN

Introduction: Electromyography (EMG) analysis is one of the most fundamental approaches for diagnosing neuromuscular diseases. Current advancements in technology have the potential to improve diagnosis accuracy using artificial intelligence (AI). Quantum machine learning (QML), while still in its early stages, offers promising potential for various medical applications, but its effectiveness in real-world diagnostic tasks needs further exploration. Thus, the aim of this study is to employ both quantum and classical support vector machines (SVMs) to classify EMG signals into two classes, healthy and myopathy, and compare their performance. Methods: Various approaches were tested; classical SVM and quantum-kernel-based SVM, both with manually extracted features, and convolutional neural network (CNN)-based deep features extraction techniques. This allows for an evaluation of the strengths and limitations of this new technology, acknowledging the potential of both classical and quantum methods. Results: The obtained results showed that the proposed quantum methods yielded promising outcomes and comparable to classical methods. Particularly, the competitive results of the quantum SVM (QSVM) with the CNN-based deep feature extraction approach, which delivered a high training and testing accuracies of up to 96.7% and 85.1%, respectively. Conclusion: These findings encourages the necessity for more advanced QML research, particularly in medical applications as quantum technology progresses.

2

Application of machine learning in management of water resources systems : a case study

Panovska Georgievska Frosina, Mitovski Stevcho, Petkovski Ljupcho

Journal of Water and Land Development

|

2025

|

Special Issue

1--12

EN

The use of machine learning (ML) in water resources management has grown due to its ability to process large, nonlinear datasets and generate accurate predictive models. In dam engineering and reservoir operation, where multiple interacting variables influence decision-making, ML provides a powerful alternative to traditional methods. The study in question investigates the application of six ML models - artificial neural networks (ANNs), support vector machines/support vector regression (SVM/SVP), random forest (RF), decision trees (DT), Gaussian process regression (GPR) and boosted regression trees (BT) - to model and predict the inflow into Kalimanci reservoir, the largest and most downstream reservoir in a complex cascade system on the Bregalnica River, North Macedonia. Using simulated operational data from HEC-ResSim for a 30-year period (2021-2049), the models were trained and tested on multiple input variables, including inflows, water supply, irrigation, water level fluctuations, and hydropower parameters. Model evaluation was based on mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination (R2). The RF, DT, and BT models outperformed others, achieving R2 values of 0.999-1.000 with minimal error rates. In contrast, the GPR model was excluded due to poor accuracy. Feature importance analysis revealed that inflows from upstream reservoirs, particularly Loshana and Razlovci, were the most influential predictors. The results confirm that ensemble-learning methods offer high accuracy and reliability in modelling complex water resources systems. These models are recommended for integration into real-time operations to improve reservoir management and optimize water allocation decisions.

3

Machine learning algorithms and geographic information system techniques to predict land suitability maps for wheat cultivation in the Central Anatolia Region

Aljanabi Firas K., Dedeoğlu Mert

Journal of Ecological Engineering

|

2025

|

Vol. 26, nr 1

373--387

EN

Maintaining food security through increased agricultural production is a major concern for decision-makers, especially in areas with arid and semi-arid climatic conditions and limited natural resources. Land suitability prediction for cultivating strategic crops, including wheat, has emerged as a crucial subject for academics, decision-makers, and economists to ensure the sustainability of natural resources. This paper aims to use three soil morphological parameters, three soil physical parameters, four soil chemical parameters, and a long-term remote sensing index as input factors to produce land suitability maps for wheat cultivation based on five machine learning algorithms (MLAs): ANN, KNN, RF, SVM, and XgbTree, in the Gozlu agricultural enterprise, which is located in a semi-arid region of the Central Anatolian Plateau. To achieve this target, an inventory of 238 appropriateness points for cultivated wheat has been executed over five years, from 2019 to 2023. The outcomes revealed that the soil texture and soli available water capacity parameters were the most influential in land suitability prediction. The best performance among the MLAs was achieved by the XgbTree algorithm, which had an accuracy of 0.98 and a kappa coefficient of 0.81. Additionally, the area under the curve (AUC) was 0.90 according the receiver operating characteristics (ROC) curve approach. The results of the study demonstrated an excellent ability of the MLAs to predict land suitability for wheat cultivation in semi-arid climate conditions. This approach can play a significant role in ensuring food security and serves as an important tool for decision-makers in sustainable development. However, we propose that the approach should be examined in comparable climatic conditions with diverse crops to ensure it is a viable solution with widely cases.

4

Comparison of Machine-Learning Algorithms for SPOT 7 Multispectral Image Classification

Morale Davide, Parente Claudio, Bolognesi Salvatore Falanga

Geomatics and Environmental Engineering

|

2025

|

Vol. 19, no. 2

71--97

EN

Precise and timely land-cover identification plays an important role in effective environmental monitoring and land management. This study compares theperformanceoffive machine-learningclassifiers –supportvectormachine (SVM), decision tree (DT), normal Bayes (NB), random forest (RF), and k-nearest neighbor (k-NN) – in the land-cover mapping of the Agro Nocerino Sarnese area (Southern Italy) using high-resolution SPOT 7 pan-sharpened multispectral images with a pixel size of 1.5 m × 1.5 m. The data set consisted of blue, green, red, and near-infrared (NIR) bands and was processed with Orfeo ToolBox (OTB) software. Two data sets were analyzed: DS-3B (which included only the visible bands [blue, green, and red]), and DS-4B (which also included the NIR band). A comparison of the classifiers’ performances across various land-cover classes was conducted in order to assess their respective classification accuracy. The results showed that SVM and k-NN achieved the highest overall accuracy levels (93% and 92%, respectively) using only the visible bands, whereas the decision tree classifier performed best when the NIR band was included. Random forest achieved excellent accuracy in vegetation classes (88–99%) but struggled with misclassifications in bare soil and man-made classes such as buildings and roads. These results emphasized the significant impact of data set characteristics on classifier performance as well as the importance of band selection and pan-sharpening techniques in high-resolution land-cover mapping.

5

Prediction of soil liquefaction using machine learning approaches

Rahman Mahabub, Hossain Belal, Sayed Abu

Engineering Transactions

|

2025

|

Vol. 73, iss. 2

257--277

EN

In the geotechnical engineering field, the assessment of liquefaction potential is a critical aspect of site evaluation. This work focuses on the application of support vector machines (SVM) to improve the accuracy of liquefaction potential evaluation. Input data were collected from the authors’ previous study and include parameters such as groundwater table (GWT), depth, fineness content, peak ground acceleration (PGA), corrected SPT-N value, total stress, and effective overburden stress. Radial basis function (RBF), linear, polynomial, and sigmoid are the four SVM kernel functions that are examined in this study to model liquefaction-related data using three approaches: grid search cross-validation, k-fold cross-validation, and fuzzy c-clustering means (FCM). Several performance metrics, including accuracy, precision, recall, and the area under the receiver operating characteristics (ROC) curve (AUC), among others, are used to evaluate the developed machine learning (ML) models. The linear and polynomial functions, for the grid search cross-validation approach, show higher performance with an accuracy of 94.64%, recall of 95.55%, F1-score of 96.63, and AUC of 0.99 on the testing data. For the k-fold partitioning approach, the RBF yields higher performance metrics compared to the other three functions, with an accuracy of 92.73%, precision of 100%, F1-score of 95.0%, and AUC of 0.98. In the FCM technique, the linear and polynomial kernels again yield greater accuracy, precision, F1-score, and specificity, while, the AUC values of the sigmoid and RBF kernels are higher. The current analysis recommends the RBF over other mathematical functions based on the k-fold partitioning technique after evaluating all performance matrices.

6

Real-Time Fault Monitoring Method for Logistics Vehicles Based on Chaotic Ant Colony Algorithm

Zhang Yinxia, Wang Liang

Eksploatacja i Niezawodność

|

2025

|

Vol. 27, no. 4

art, no. 203395

EN

To improve the safety of logistics vehicle transportation, this study proposes a real-time fault monitoring method for logistics vehicles based on chaotic ant colony algorithm. Firstly, take a typical engine malfunction as an example. Identify fault signals based on logistics vehicle fault tree. Then, use support vector machine theory to extract time-domain low dimensional features from vehicle fault information. Finally, real-time fault monitoring of logistics vehicles is achieved based on chaotic ant colony optimization algorithm. The experiment shows that the monitoring accuracy of this method is always above 94.0%, and the monitoring signal transmission delay varies between 444ms-627ms, indicating that this method has high monitoring accuracy and efficiency, and has high application value.

7

Research on the construction and application of electrical fault classification system based on Bayesian algorithm

Zhang Xiaowei, Zhu Mingdong

Diagnostyka

|

2025

|

Vol. 26, No. 2

art. no. 2025213

EN

With the rapid development of smart building technology, the safe and stable operation of the electrical system has become the core demand of modern building management. This study aims to construct an automated classification system for electrical faults based on Bayesian algorithm to improve the accuracy and efficiency of fault diagnosis. First, the wavelet transform is utilized for noise reduction and feature extraction of electrical signals to enhance the signal-to-noise ratio of the data. Subsequently, multi-category fault diagnosis is realized based on association vector machine, and Bayesian approach is combined to quantify the uncertainty factors and improve the classification reliability. The results show that the system performs well with small sample data, and the average recognition accuracy of various types of faults exceeds 70%. The wavelet transform-based fault recognition method demonstrates high stability, with the highest accuracy reaching 100% and the lowest still maintaining around 90%. In addition, the Bayesian classifier significantly improves the confidence level of fault diagnosis after parameter optimization, which verifies the effectiveness of the algorithm. It provides a feasible solution for the fault prediction and health management of power systems in intelligent buildings.

8

Power transformer fault warning combining support vector machine and improved grey wolf optimization algorithm

Zhao Shuzong, Pattanadech Norasage

Archives of Electrical Engineering

|

2025

|

Vol. 74, nr 1

191--208

EN

To optimize the parameter setting of the support vector machine and improve the classification performance and computational efficiency of power transformer fault diagnosis, this study proposes an improved grey wolf optimization algorithm. By optimizing the global search and local optimization capabilities of the grey wolf algorithm and combining them with stacked denoising autoencoders, a new power transformer fault warning model is constructed. Firstly, the grey wolf optimization algorithm is optimized through four strategies: elite reverse learning, nonlinear control parameters, Lévy flight, and particle swarm optimization, which improve its global search and local optimization capabilities. Secondly, the stacked denoising autoencoder is utilized to extract high-level features of fault data, and the improved GWO algorithm and SVM are combined to complete fault classification. The results indicated that the proposed diagnostic model achieved a diagnostic accuracy of 0.979, a recall rate of 0.986, and an F1 value of 0.983 in benchmark performance testing. In practical applications, the average fault diagnosis accuracy of this model could reach up to 99.21%, and the average diagnosis time was only 0.08 s. The developed power transformer fault warning model can provide an efficient and reliable technical solution for fault diagnosis in the power system.

9

Detecting clustered fruits using a hybrid of convolutional neural networks and machine learning classifiers – Case study

Falih Bashar S., Gierz Łukasz, Al-Zaidi Ghazwan A.

Advances in Science and Technology. Research Journal

|

2025

|

Vol. 19, no. 4

1--9

EN

The last and most important procedure during fruit or vegetable cultivation is harvesting. One of the basic challenges during grape growing is the use of agriculture 4.0 machines (including robots) during harvesting which is associated with the need for quick identification of berries or grape clusters. In this work, a convolutional neural network (CNN) and a machine learning classifier were suggested for the identification (detection) of individual grapes. A free data set (Iceland) was used, which included two classes with different lighting conditions and berry sizes. The integrated method included two types of deep learning models, i.e. CNN (AlexNet and GoogleNet). CNN models were used to obtain discriminative deep features from different layers. The combination of two models AlexNet-Fc6 and SVM-Cubic yielded the highest accuracy, sensitivity and precision (mean ± standard deviation) % of 99.4 ± 0.13, 99.2 ± 0.14 and 99.49 ± 0.19, respectively. The developed grape detector can be used for practical applications requiring high accuracy, e.g. in the process of yield estimation or detection of grape diseases.

10

Comparison of machine learning models for predicting the compressive strength of cement mixtures with zeolite

Michaluk Justyna, Kulisz Monika, Kujawska Justyna, Wojtaś Edyta, Aldungarova Aliya

Advances in Science and Technology. Research Journal

|

2025

|

Vol. 19, no. 10

123--135

EN

This study investigates the applicability of machine learning algorithms for predicting the compressive strength of cement mixtures with zeolite. The research compares the performance of four predictive models—Elastic Net regression, Support Vector Machines (SVM), Multilayer Perceptron (MLP) neural networks, and Decision Trees—trained on experimentally obtained data describing mix composition and curing conditions. The input features included zeolite percentage, water-to-cementitious-material ratio, curing time, cement mass, and zeolite content. The output variable was compressive strength. Among the evaluated models, the SVM algorithm exhibited the optimal generalization capability, attaining the minimal prediction error on the validation set while sustaining elevated correlation between actual and predicted values. The MLP neural network demonstrated the optimal fit to the training data, however, this was achieved at the expense of heightened sensitivity to overfitting. Decision trees demonstrated robust training efficacy but exhibited diminished generalization capabilities, while the linear elastic net model encountered challenges in replicating the nonlinear characteristics of the material system. The study corroborates the viability of nonlinear machine learning models in facilitating the design and optimization of zeolite-enhanced cementitious mixtures. These findings signify a significant stride towards data-driven modeling in the field of construction materials engineering, thereby facilitating enhanced prediction of mechanical performance with minimized experimental effort. The study also underscores avenues for future exploration, encompassing model hybridization, multi-output prediction frameworks, and integration with optimization algorithms for automated mix design.

11

Global-scale explainable AI assessment for OBIA-based classification using Deep Learning and Machine Learning methods

Kavzoglu Taskin, Uzun Yakup Kaan, Berkan Emre, Yilmaz Elif Ozlem

Advances in Geodesy and Geoinformation

|

2025

|

Vol. 74, no. 1

art. no. e62, 2025

EN

Over the past decade, object-based image analysis (OBIA) has gained prominence as a widely adopted method for generating land use/land cover (LULC) maps. This study aims to evaluate the performance of various classification algorithms within the OBIA framework using SPOT-6 satellite imagery. The research methodology involved segmenting the images with the multi-resolution segmentation (MRS) algorithm, followed by the application of convolutional neural networks (CNN), random forest (RF), and support vector machine (SVM) algorithms for classification. The study was conducted in the Perpignan province, located in the Pyrénées-Orientales region of France. After the segmentation stage, CNN, RF, and SVM classifiers were employed to classify the image segments based on both spectral and spatial attributes. The accuracy of the resulting thematic maps was assessed using standard metrics, including overall accuracy (OA), the Kappa coefficient (KC), and the F��score (FS). Of the three classifiers, CNN achieved the highest overall accuracy at 91.28%, outperforming SVM, which attained an OA of 90.50%, and RF, which recorded an OA of 87.28%. Additionally, this study explored the integration of explainable artificial intelligence (AI) techniques, specifically the Shapley Additive Explanations (SHAP) algorithm, to enhance the interpretability of the machine learning models. This approach fosters greater trust, accountability, and acceptance in decision-making processes. By leveraging SHAP values, the study provides deeper insights into the decision-making processes of the CNN, SVM, and RF classifiers, ultimately enhancing the transparency and comprehensibility of these models.

12

Support vector machine to criminal recidivism prediction

Kovalchuk Olha, Shevchuk Ruslan, Babala Ludmila, Kasianchuk Mykhailo

International Journal of Electronics and Telecommunications

|

2024

|

Vol. 70, No. 3

691-697

EN

Internal security of the state is one of the prerequisites for sustainable development. To ensure the public safety and personal security of citizens, it is necessary to develop effective measures to reduce crime and prevent crime in the future. The starting point for the development and practical implementation of an effective strategy to combat crime or prevent certain crimes is criminological forecasting. Individual forecasting is aimed at determining the possibility of committing a crime (crimes) in the future by a certain person or group of persons. For risk assessment, the following are traditionally used machine learning models. Such models also provide qualitative assessments in the scientific prediction of the likelihood and possibilities of committing a repeat criminal offense. Knowledge gained from the application of machine learning algorithm, can provide justice authorities with anticipatory information that is essential for developing a general concept of combating crime. The development of applied models for crime analysis and forecasting can become a reliable tool to support decision-making in predicting likely criminal behavior in the future and ensuring the internal security of the state. In this paper, the results of the application are presented by the machine-learning algorithms Support Vector Machine (SVM) for assessment of the risk of recidivism of criminal offenses by persons who have already been convicted of such a crime in the past. The data set consisted of the 12,000 criminal defendants’ criminal profile information in Ukraine. The constructed classifier has a high precision (98.67%), recall (97.53%) and is qualitative (AUC is equal 0.981). The created SVM model can be applied to new data set to predict the risk of reoffending by convicted individuals in the future.

13

Support Vector Machine for Susceptibility Modeling of Dengue Fever in Kendari, Southeast Sulawesi

Widayani Prima, Sahity Abhista Fawwaz, Saputri Agatha Andriantari

Geomatics and Environmental Engineering

|

2024

|

Vol. 18, no. 1

29--50

EN

Dengue fever (DF) is an infectious disease that is still a problem in Indonesia. The total death rate due to DF was 705 people in 2021; in 2022, this increased to 1183 (Indonesian Ministry of Health, 2023). Seeing this fact, prevention efforts are still needed when handling DF cases in all of the regions of Indonesia. This research was conducted in the Kendari area of Southeast Sulawesi, where there are still cases of DF. The purpose of this study was to create a spatial model of dengue susceptibility using a support vector machine. Landsat 8 imagery was used to intercept data on building density, vegetation density, land use, and land surface temperatures. Rainfall and humidity variables were obtained from the Meteorological, Climatological, and Geophysical Agency (BMKG). Based on the modeling results, the districts of Wua-wua, Kadia, Barunga, Poasi, and Puuwatu are areas with high susceptibility. The results of testing the susceptibility model to dengue hemorrhagic fever (DHF) in Kendari obtained an area under the curve (AUC) of 0.75, meaning that this model was well-accepted.

14

Boosting and Comparing Performance of Machine Learning Classifiers with Meta-heuristic Techniques to Detect Code Smell

Jain Shivani, Saha Anju

e-Informatica Software Engineering Journal

|

2024

|

Vol. 18, nr 1

240107

EN

Background: Continuous modifications, suboptimal software design practices, and stringent project deadlines contribute to the proliferation of code smells. Detecting and refactoring these code smells are pivotal to maintaining complex and essential software systems. Neglecting them may lead to future software defects, rendering systems challenging to maintain, and eventually obsolete. Supervised machine learning techniques have emerged as valuable tools for classifying code smells without needing expert knowledge or fixed threshold values. Further enhancement of classifier performance can be achieved through effective feature selection techniques and the optimization of hyperparameter values. Aim: Performance measures of multiple machine learning classifiers are improved by fine tuning its hyperparameters using various type of meta-heuristic algorithms including swarm intelligent, physics, math, and bio-based etc. Their performance measures are compared to find the best meta-heuristic algorithm in the context of code smell detection and its impact is evaluated based on statistical tests. Method: This study employs sixteen contemporary and robust meta-heuristic algorithms to optimize the hyperparameters of two machine learning algorithms: Support Vector Machine (SVM) and k-nearest Neighbors (k-NN). The No Free Lunch theorem underscores that the success of an optimization algorithm in one application may not necessarily extend to others. Consequently, a rigorous comparative analysis of these algorithms is undertaken to identify the best-fit solutions for code smell detection. A diverse range of optimization algorithms, encompassing Arithmetic, Jellyfish Search, Flow Direction, Student Psychology Based, Pathfinder, Sine Cosine, Jaya, Crow Search, Dragonfly, Krill Herd, Multi-Verse, Symbiotic Organisms Search, Flower Pollination, Teaching Learning Based, Gravitational Search, and Biogeography-Based Optimization, have been implemented. Results: In the case of optimized SVM, the highest attained accuracy, AUC, and F-measure values are 98.75%, 100%, and 98.57%, respectively. Remarkably, significant increases in accuracy and AUC, reaching 32.22% and 45.11% respectively, are observed. For k-NN, the best accuracy, AUC, and F-measure values are all perfect at 100%, with noteworthy hikes in accuracy and ROC-AUC values, amounting to 43.89% and 40.83%, respectively. Conclusion: Optimized SVM exhibits exceptional performance with the Sine Cosine Optimization algorithm, while k-NN attains its peak performance with the Flower Optimization algorithm. Statistical analysis underscores the substantial impact of employing meta-heuristic algorithms for optimizing machine learning classifiers, enhancing their performance significantly. Optimized SVM excels in detecting the God Class, while optimized k-NN is particularly effective in identifying the Data Class. This innovative fusion automates the tuning process and elevates classifier performance, simultaneously addressing multiple longstanding challenges.

15

Construction and application of a bearing fault diagnosis model based on improved GWO algorithm

Jiang Lingbo

Diagnostyka

|

2024

|

Vol. 25, No. 3

art. no. 2024302

EN

In mechanical equipment, if bearing components fail, it can cause serious equipment damage and even threaten human life safety. Therefore, equipment bearings fault diagnosis is of great meaning. In the study of bearing fault diagnosis, an improved gray wolf optimization algorithm is put forward to optimize the support vector machine model. The model improves the convergence factor of the algorithm, and then optimizes the penalty factor and KF parameters of the support vector machine to enhance the accuracy of fault classification. At the same time, in the problem of fault identification, the introduction of adaptive noise set empirical mode decomposition and the combination of permutation entropy are studied to minimize the impact of noise on the identification model. The experimental outcomes indicated that the algorithm proposed in the study had an average fitness value and a standard deviation fitness value of 0 in the benchmark test function and 94.55% accuracy in overall fault identification. The permutation entropy of the vibration signal in the normal state of the bearing was within the range of [0.13, 0.52], which has a more stable state compared to the fault state. The results show that the improved grey Wolf optimization algorithm is applied to the optimization of support vector machine, combined with the adaptive noise set empirical mode decomposition and permutation entropy, and the identification and classification results of bearing faults are successfully improved, making the proposed method feasible in bearing fault diagnosis, and providing a more effective scheme for fault diagnosis.

16

Advanced AI tools for predicting mechanical properties of self-compacting concrete

Agrawal Achal, Chandak Narayan

Architecture Civil Engineering Environment

|

2024

|

Vol. 17, no. 2

69--86

EN

The present study utilizes advanced numerical evaluation techniques like Artificial Intelligence (AI), including Support Vector Machines (SVM), Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems with Genetic Algorithms (ANFIS-GA), Gene Expression Programming (GEP), and Multiple Linear Regression (MLR) to develop and compare the predictive models for determination of compressive and tensile strength. Partial mutual information for selection and establishment of the degree of association of variables was used to aid in better attainment of results obtained through predictive models. It was observed that amongst the modeling techniques, the results obtained for compressive strength through the SVM technique were excellent, producing an Index of Agreement of 0.96, Akaike Information Criterion of 68.33, skill score of 0.96, and symmetric uncertainty of 0.93, thus indicating a simpler, robust, and low uncertainty predictive model. Furthermore, the adapted technique MLR was found to predict tensile strength characteristics better, with the MLR model demonstrating a higher R2 value of 0.81, thus implying a reliable tensile strength prediction model. However, SVM consistently performed well for both compressive and tensile strength characteristics thus endorsing the reliability of the predictive model. Overall, the study aids in getting new insights about improvising the strength properties of SCC and its evaluation through predictive techniques.

17

Emotion recognition from heart rate variability using a hybrid system combined with a Hidden Markov model and Poincare plot

Khanghah Sahar Zamani, Maghooli Keivan

Applied Computer Science

|

2024

|

Vol. 20, no. 1

106--121

EN

The best emotion recognition system based on physiological signals with a simple operatory should have higher accuracy and fast response speed. This paper aims to develop an emotion recognition system using a novel hybrid system based on Hidden Markov Model and Poincare plot. For this purpose, an electrocardiogram from the MAHNOB-HCI database was used. A novel feature extraction from a hybrid system combining Hidden Markov Model and Poincare plot was presented. The authors extracted time and frequency domain features from heart rate variability, and used two central hybrid systems, the Support Vector Machine/ Hidden Markov Model and the Hidden Markov Model/ Poincare Plot. Finally, the support vector machine was used as a classifier to classify emotions into positive and negative. The proposed method showed a classification accuracy of 95.02 ± 1.97 % overall. Also, the computing time of the method is around 163 milliseconds. The key of this paper is in the use of hybrid machines to improve accuracy without high computation time. This method can be used as a real-time system due to the low computation time and can be developed in many fields, such as medical examination and security systems.

18

Modeling daily reference evapotranspiration using SVR machine learning algorithm with limited meteorological data in Dar-el-Beidha, Algeria

Salah Zereg, Khaled Belouz

Acta Geophysica

|

2024

|

Vol. 72, no. 3

2009--2025

EN

Reference evapotranspiration (ETo) is a critical water resource management parameter, including irrigation scheduling and crop water requirements. Because large uncertainties in estimating ETo can result in equally large uncertainties in determining water budgets and crop water requirements, and vice versa, accurate determination of ETo can be challenging when direct measurement and estimation with the Penman-Monteith (FAO-56-PM) semi-empirical equation of the food and agriculture organization (FAO) is not possible. Indeed, this study explores the use of the support vector regression machine learning algorithm (SVR) to predict daily ETo with limited measured inputs. It is the first time that Julian Day (J) is included as an input to improve prediction accuracy. Ten years of meteorological data collected at the Dar-El-Beidha weather station in Algeria are used, with maximum, minimum, and mean air temperatures (TM, tm, and T), mean relative humidity (RH), mean wind speed (u2), and sunshine duration (n) as inputs, as well as J and extraterrestrial solar radiation (Ra) as auxiliary variables, and the ETo-FAO-56-PM values as target outputs. Several SVR models are developed using different combinations of inputs, and their performance is assessed relative to ETo-FAO-56-PM values. Empirical equations are also used for comparison, and several evaluation metrics are employed, including root mean square error (RMSE), mean absolute percentage error (MAPE), determination coefficient (R2), RMSE-standard deviation ratio (RSR), Nash-Sutcliffe efficiency coefficient (NSE), and Willmott’s refined index (WI). The results show that the SVR models utilizing limited meteorological inputs in addition to J and/or Ra predicted ETo accurately and outperformed their corresponding estimates using empirical equations, radial basis function neural networks (RBFNN), and adaptive neuro-fuzzy inference systems (ANFIS) models obtained in previous studies. The RMSE ranged from 0.28 to 0.72 mm/day, R2 from 0.86 to 0.98, MAPE from 7 to 19%, RSR from 0.15 to 0.38, NSE from 0.86 to 0.98, and WI from 0.65 to 0.87. These findings could provide useful solutions for ETo estimation issues in areas with sparse data and agro-climatic conditions similar to those of Dar-El-Beidha.

19

High-resolution mapping of seasonal snow cover extent in the Pamir Hindu Kush using machine learning-based integration of multi-sensor data

Mahmoodzada Abdul Basir, Das Pragyan, Varade Divyesh, Akhtar Mohd Arslaan, Shimada Sawahiko

Acta Geophysica

|

2024

|

Vol. 72, no. 2

1455--1470

EN

This study proposes a framework to develop a high-resolution snow cover area (SCA) product from freely available spaceborne remote sensing data and utilizes the Sentinel-1 multi-temporal products and MODIS surface reflectance data. The proposed methodology focuses on using the sensitivity of the parameters retrievable from the Sentinel-1 datasets to snow. Different parameters such as the dual polarimetric entropy, mean scattering angle, backscatter coefficients, and the interferometric coherence are integrated with a spatially resampled normalized difference snow index (NDSI) from MODIS data to estimate an equivalent NDSI, which is used for the determination of the SCA at 15 m spatial resolution. The equivalent NDSI is derived using a machine learning-based regression based on support vector machines (SVMs) and the multilayer perceptron (MLP). The experiments are performed for the high elevated regions of the Kunduz and Khanabad watershed of the northern Hindu Kush mountains for the peak winter and early melt season of 2019, corresponding to February and March. The reference SCA for evaluating the results is generated by thresholding the NDSI derived from pan-sharpened Landsat-8 imagery. As compared to MLP, the SCA generated based on the SVM regression showed better performance. Further, compared to spatially resampled MODIS NDSI, both the SVM and MLP results showed better accuracy for snow classification, as determined by the mean conditional kappa coefficients of 0.75, 0.83, respectively, over 0.62.

20

Continuous speech commands recognition with thai language used support vector machine technique: A case study of speech commands control for mobile robots

Muangjaroen Supavit, Udomsiri Sakol

Przegląd Elektrotechniczny

|

2023

|

R. 99, nr 6

181--187

EN

This article presents the application of an automatic speech recognition by continuous speech commands recognition with Thai language as a speaker verification model, this is a case study of speech commands control of mobile robots. The design of the automatic speech recognition system consisted of 3 steps: The first we analyzed the signal processing of the continuous speech commands and compared the accuracy of the speech recognition with a time frame adjustment and the overlapped period of signal filtered with the window function, The second we proceed to find the feature extraction of speech commands using format frequency techniques and configured the feature extraction with format frequencies of F1, F2, and F3,The last step was to design the recognition using Support Vector Machine technique to check the accuracy of an automatic speech recognition. These is support vector machine classification algorithm provides a comparison of the filtered function window and compares the accuracy of the time frame scaled and the overlapped time of the filtered, which gives different values of precision. From the experiment, the researcher found that are applied a Hanging function the test results of the test result of the "forward" speech commands has an accuracy of 81.92% but kind of Gaussian function the test results of the "backward" speech commands has an accuracy of 83.69%, the "turn left" speech commands had an accuracy of 82.81%, the "turn right" speech commands had an accuracy of 85.56% and the "Stop first" speech commands has an accuracy of 86.78% and speech recognition by continuous speech commands recognition with Thai language was applied an every function the test results of the overall performance of the speech commands has an accuracy of 83.88%.

PL

Artykuł przedstawia zastosowanie automatycznego rozpoznawania mowy poprzez ciągłe rozpoznawanie poleceń głosowych z językiem tajskim jako modelem weryfikacji mówiącego, jest to studium przypadku sterowania poleceniami głosowymi robotów mobilnych. Projekt systemu automatycznego rozpoznawania mowy składał się z 3 etapów: W pierwszym przeanalizowano przetwarzanie sygnału ciągłych poleceń głosowych i porównano dokładność rozpoznawania mowy z dopasowaniem przedziału czasowego i nakładającym się okresem sygnału filtrowanego funkcją okna, Następnie przystępujemy do znalezienia ekstrakcji funkcji poleceń głosowych przy użyciu technik formatowania częstotliwości i skonfigurowania ekstrakcji cech z częstotliwościami formatu F1, F2 i F3. Ostatnim krokiem było zaprojektowanie rozpoznawania przy użyciu techniki maszyny wektorów nośnych w celu sprawdzenia dokładności automatyczne rozpoznawanie mowy. Jest to algorytm klasyfikacji maszyny wektorów nośnych, który zapewnia porównanie przefiltrowanego okna funkcji i porównuje dokładność skalowanych ram czasowych oraz nakładających się czasów filtrowanych, co daje różne wartości precyzji. Na podstawie eksperymentu badacz odkrył, że po zastosowaniu funkcji wiszącej wyniki testu wyników poleceń głosowych „do przodu” mają dokładność 81,92%, ale rodzaj funkcji Gaussa wyniki testu poleceń głosowych „wstecz” mają dokładność 81,92% dokładność 83,69%, polecenia głosowe „skręć w lewo” miały dokładność 82,81%, polecenia głosowe „skręć w prawo” miały dokładność 85,56%, a polecenia głosowe „Najpierw zatrzymaj” mają dokładność 86,78%, a rozpoznawanie mowy przez zastosowano ciągłe rozpoznawanie poleceń głosowych w języku tajskim, a wyniki testu ogólnej wydajności poleceń głosowych mają dokładność 83,88%.