Basis Function Kernel Support Vector Research Articles

Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer

BackgroundPathological complete response (pCR) is associated with favorable prognosis in patients with triple-negative breast cancer (TNBC). However, only 30–40% of TNBC patients treated with neoadjuvant chemotherapy (NAC) show pCR, while the remaining 60–70% show residual disease (RD). The role of the tumor microenvironment in NAC response in patients with TNBC remains unclear. In this study, we developed a machine learning-based two-step pipeline to distinguish between various histological components in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of TNBC tissue biopsies and to identify histological features that can predict NAC response.MethodsH&E-stained WSIs of treatment-naïve biopsies from 85 patients (51 with pCR and 34 with RD) of the model development cohort and 79 patients (41 with pCR and 38 with RD) of the validation cohort were separated through a stratified eightfold cross-validation strategy for the first step and leave-one-out cross-validation strategy for the second step. A tile-level histology label prediction pipeline and four machine-learning classifiers were used to analyze 468,043 tiles of WSIs. The best-trained classifier used 55 texture features from each tile to produce a probability profile during testing. The predicted histology classes were used to generate a histology classification map of the spatial distributions of different tissue regions. A patient-level NAC response prediction pipeline was trained with features derived from paired histology classification maps. The top graph-based features capturing the relevant spatial information across the different histological classes were provided to the radial basis function kernel support vector machine (rbfSVM) classifier for NAC treatment response prediction.ResultsThe tile-level prediction pipeline achieved 86.72% accuracy for histology class classification, while the patient-level pipeline achieved 83.53% NAC response (pCR vs. RD) prediction accuracy of the model development cohort. The model was validated with an independent cohort with tile histology validation accuracy of 83.59% and NAC prediction accuracy of 81.01%. The histological class pairs with the strongest NAC response predictive ability were tumor and tumor tumor-infiltrating lymphocytes for pCR and microvessel density and polyploid giant cancer cells for RD.ConclusionOur machine learning pipeline can robustly identify clinically relevant histological classes that predict NAC response in TNBC patients and may help guide patient selection for NAC treatment.

Surface Enhanced Raman Spectroscopy Pb2+ Ion Detection Based on a Gradient Boosting Decision Tree Algorithm

Lead pollution poses a serious threat to the natural environment, and a fast and high-sensitivity method is urgently needed. SERS can be used for the detection of Pb2+ ions, which is urgently needed. Based on the SERS spectral reference data set of lead nitride (Pb(NO3)2), a model for detecting Pb2+ was established by using a traditional machine learning algorithm and the GBDT algorithm. Principal component analysis was used to compare the batch effect reduction in different pretreatment methods in order to find the optimal combination of such methods and machine learning models. The combination of LightGBM algorithms successfully identified Pb2+ from cross-batch data, exceeding the 84.6% balanced accuracy of the baseline correction+ radial basis function kernel support vector machine (BC+RBFSVM) model and showing satisfactory results, with a 91.4% balanced accuracy and a 0.9313 area under the ROC curve.

Comparing Algorithms for Estimation of Aboveground Biomass in Pinus yunnanensis

Comparing algorithms are crucial for enhancing the accuracy of remote sensing estimations of forest biomass in regions with high heterogeneity. Herein, Sentinel 2A, Sentinel 1A, Landsat 8 OLI, and Digital Elevation Model (DEM) were selected as data sources. A total of 12 algorithms, including 7 types of learners, were utilized for estimating the aboveground biomass (AGB) of Pinus yunnanensis forest. The results showed that: (1) The optimal algorithm (Extreme Gradient Boosting, XGBoost) was selected as the meta-model (referred to as XGBoost-stacking) of the stacking ensemble algorithm, which integrated 11 other algorithms. The R2 value was improved by 0.12 up to 0.61, and RMSE was decreased by 4.53 Mg/ha down to 39.34 Mg/ha compared to the XGBoost. All algorithms consistently showed severe underestimation of AGB in the Pinus yunnanensis forest of Yunnan Province when AGB exceeded 100 Mg/ha. (2) XGBoost-Stacking, XGBoost, BRNN (Bayesian Regularized Neural Network), RF (Random Forest), and QRF (Quantile Random Forest) have good sensitivity to forest AGB. QRNN (Quantile Regression Neural Network), GP (Gaussian Process), and EN (Elastic Network) have more outlier data and their robustness was poor. SVM-RBF (Radial Basis Function Kernel Support Vector Machine), k-NN (K Nearest Neighbors), and SGB (Stochastic Gradient Boosting) algorithms have good robustness, but their sensitivity was poor, and QRF algorithms and BRNN algorithm can estimate low values with higher accuracy. In conclusion, the XGBoost-stacking, XGBoost, and BRNN algorithms have shown promising application prospects in remote sensing estimation of forest biomass. This study could provide a reference for selecting the suitable algorithm for forest AGB estimation.

Mind wandering state detection during video-based learning via EEG.

The aim of this study is to explore the potential of technology for detecting mind wandering, particularly during video-based distance learning, with the ultimate benefit of improving learning outcomes. To overcome the challenges of previous mind wandering research in ecological validity, sample balance, and dataset size, this study utilized practical electroencephalography (EEG) recording hardware and designed a paradigm consisting of viewing short-duration video lectures under a focused learning condition and a future planning condition. Participants estimated statistics of their attentional state at the end of each video, and we combined this rating scale feedback with self-caught key press responses during video watching to obtain binary labels for classifier training. EEG was recorded using an 8-channel system, and spatial covariance features processed by Riemannian geometry were employed. The results demonstrate that a radial basis function kernel support vector machine classifier, using Riemannian-processed covariance features from delta, theta, alpha, and beta bands, can detect mind wandering with a mean area under the receiver operating characteristic curve (AUC) of 0.876 for within-participant classification and AUC of 0.703 for cross-lecture classification. Furthermore, our results suggest that a short duration of training data is sufficient to train a classifier for online decoding, as cross-lecture classification remained at an average AUC of 0.689 when using 70% of the training set (about 9 min). The findings highlight the potential for practical EEG hardware in detecting mind wandering with high accuracy, which has potential application to improving learning outcomes during video-based distance learning.

Detecting paroxysmal atrial fibrillation from normal sinus rhythm in equine athletes using Symmetric Projection Attractor Reconstruction and machine learning

BackgroundAtrial fibrillation (AF) is a common cardiac arrhythmia in both human and equine populations. It is associated with adverse outcomes in humans and decreased athletic performance in both populations. Paroxysmal atrial fibrillation (PAF) presents with intermittent, self-terminating AF episodes, and is difficult to diagnose once sinus rhythm resumes.ObjectiveWe aimed to detect PAF subjects from normal sinus rhythm equine electrocardiograms (ECGs) using the Symmetric Projection Attractor Reconstruction (SPAR) method to encapsulate the waveform morphology and variability as the basis of a machine learning classification.MethodsWe obtained ECG signals from 139 active equine athletes (120 control, 19 with a PAF diagnosis). The SPAR method was applied to 9 short (20-second) ECG strips for each subject. An optimal SPAR feature set was determined by forward feature selection for input to a machine learning model ensemble of 3 different classifiers (k-nearest neighbors, linear support vector machine, and radial basis function kernel support vector machine). Imbalanced data were handled by upsampling the minority (PAF) class. A final subject classification was made by taking a majority vote over results from the 9 ECG strips.ResultsOur final cross-validated classification for a subject gave an accuracy of 89.0%, sensitivity of 94.8%, specificity of 87.1%, and receiver operating characteristic area under the curve of 0.98, taking PAF as the positive class.ConclusionThe SPAR method and machine learning generated a final model with high sensitivity, suggesting that PAF can be discriminated from short equine ECG strips. This preliminary study indicated that SPAR analysis of human ECG could support patient monitoring, risk stratification, and clinical decision-making.

Breast Cancer Diagnosis Using Support Vector Machines Optimized by Whale Optimization and Dragonfly Algorithms

Breast Cancer (BC) has become a critical illness with a high mortality rate during the previous decade. It is considered the women’s most common cancer. In this paper, we propose two optimum automated BC classification approaches based on a hybridization of the Whale Optimization Algorithm (WOA) and Dragonfly Algorithm (DA), with Radial Basis Function Kernel Support Vector Machines (RBF-SVM), to increase the accuracy of BC classification (CA) by determining the optimum SVM parameters. The effectiveness of the proposed WOA-SVM and DA-SVM algorithms is tested on the Wisconsin Diagnosis Breast Cancer (WDBC) databases and the Wisconsin Breast Cancer Database (WBCD). Various metric parameters such as CA, confusion matrix, the area under the ROC curve (AUC), sensitivity, and specificity are utilized to assess and consider the effectiveness of the proposed approaches. The results are compared not only to the most common optimizers, Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), that are used for training SVM and artificial neural networks (ANN) classifiers, but also to other classification models. The WOA-SVM and DA-SVM are also explored for feature selection, and their findings are compared to the offered models. According to the experimental results, the proposed WOA-SVM method outperforms previous classification approaches on the WBCD dataset. On the WDBC dataset, however, the proposed DA-SVM algorithm outperforms the previous classification algorithms. Using typical datasets’ partition, the resultant CA is as high as 99.65% and 100% for WDBC and WBCD, respectively. However, using a 10-fold cross-validation datasets’ partition, the mean resultant CA are 97.89% and 99.27%, respectively.

Dimensional reduction based on peak fitting of Raman micro spectroscopy data improves detection of prostate cancer in tissue specimens.

.Significance: Prostate cancer is the most common cancer among men. An accurate diagnosis of its severity at detection plays a major role in improving their survival. Recently, machine learning models using biomarkers identified from Raman micro-spectroscopy discriminated intraductal carcinoma of the prostate (IDC-P) from cancer tissue with a detection accuracy and differentiated high-grade prostatic intraepithelial neoplasia (HGPIN) from IDC-P with a accuracy.Aim: To improve the classification performance of machine learning models identifying different types of prostate cancer tissue using a new dimensional reduction technique.Approach: A radial basis function (RBF) kernel support vector machine (SVM) model was trained on Raman spectra of prostate tissue from a 272-patient cohort (Centre hospitalier de l’Université de Montréal, CHUM) and tested on two independent cohorts of 76 patients [University Health Network (UHN)] and 135 patients (Centre hospitalier universitaire de Québec-Université Laval, CHUQc-UL). Two types of engineered features were used. Individual intensity features, i.e., Raman signal intensity measured at particular wavelengths and novel Raman spectra fitted peak features consisting of peak heights and widths.Results: Combining engineered features improved classification performance for the three aforementioned classification tasks. The improvements for IDC-P/cancer classification for the UHN and CHUQc-UL testing sets in accuracy, sensitivity, specificity, and area under the curve (AUC) are (numbers in parenthesis are associated with the CHUQc-UL testing set): (), (), (6%), () with respect to the current best models. Discrimination between HGPIN and IDC-P was also improved in both testing cohorts: (), (), (), (). While no global improvements were obtained for the normal versus cancer classification task [ (), (), (), ()], the AUC was improved in both testing sets.Conclusions: Combining individual intensity features and novel Raman fitted peak features, improved the classification performance on two independent and multicenter testing sets in comparison to using only individual intensity features.

Stability analysis of distributed smart grid based on machine learning

Decentralized Smart Grid (DSG) is a new technology proposed to power networks with elastic nodes. It can realize dynamic electricity price demand response without large-scale transformation of the infrastructure.In order to analyze the system stability of DSG, six representative machine learning classification models were applied to analyze the stability data of 10,000 samples of the 4-node system. Combined with the requirements of power system security, stability and economic performance, the effect of each classification model on the stability prediction of DSG system was tested.The test results showed that the model with a Gaussian kernel basis function kernel support Vector machine (RBF SVM) was suitable for data analysis, with the accuracy up to 97.10% and the F metric up to 0.977.CART decision tree model is suitable for real-time forecasting of power system. Under the requirement of ensuring real-time forecasting, its accuracy can reach 84.90% and F metric can reach 0.882. Its modeling and prediction calculation requirements are only 0.98% and 1.59% of that of RBF SVM model respectively.

PREDICTIVE MODELLING AND ANALYTICS FOR DIABETES USING A MACHINE LEARNING APPROACH

Adverse effects can be seen in the entire body due to the major disorders known as Diabetes. The risk of dangers like diabetic nephropathy, cardiac stroke and other disorders can increase severally because of the undiagnosed diabetes. Around the globe the people are suffering from this disease. For a healthy life early detection of this disease is very curtail. As the causes of the diabetes is increasing rapidly this disease might turn up as a reason for worldwide concern. Increasing the chances for a more accurate predictions and form experiences automatic learning by computational method may be provided by Machine Learning (ML). With the help of R data manipulation tool for trends development and with risk factor patterns detection in Pima Indian diabetes technique of machine learning is been used in the current researches. With the use of R data manipulation tool analysis and development five different predictive models is done for the categorization of patients into diabetic and non- diabetic. supervised machine learning algorithms namely multifactor dimensionality reduction (MDR), k-nearest neighbor (k-NN), artificial neural network (ANN) radial basis function (RBF) kernel support vector machine and linear kernel support vector machine (SVM-linear) are used for this purpose.

Modelling heating and cooling energy demand for building stock using a hybrid approach

The building sector accounts for 30% of final energy consumption and 28% of global energy-related carbon dioxide emissions, with space heating and cooling consuming a large share of total buildings’ energy consumption. Building stock modelling for space heating and cooling energy prediction provides critical insights on the stock energy consumption and aid the building retrofit policy-making process with the evaluation of the energy-saving potential. By combining the physical modelling approach and data-driven approach, a hybrid approach is applicable for modelling the heating and cooling energy consumption of the building stock, including both residential buildings and non-residential buildings. Within this framework, the Urban Modelling Interface (UMI) tool has been used for physical modelling to generate heating and cooling energy use intensity. Then, ten different machine learning models, including Gaussian radial basis function kernel support vector regression, linear kernel support vector regression, polynomial kernel support vector regression, random forests, extreme gradient boosting, ordinary least-squares linear regression, ridge regression, least absolute shrinkage and selection operator, elastic net and artificial neural network, have been applied to predict heating and cooling energy use intensity (EUI). The approach has been demonstrated using a case study in Chongqing, China. The results show that machine learning models can achieve accurate building heating and cooling EUI prediction, with the polynomial kernel support vector regression showing the best accuracy at the level of a single building, and the Gaussian radial basis function kernel support vector regression performing the best at the stock level. Machine learning models generated by proposed hybrid approach not only provide quickly prediction of building space heating and cooling energy consumption at the stock level, but also support building retrofit decision makings by evaluate energy saving potential of various retrofit options.

A machine-learning-based approach to predict residential annual space heating and cooling loads considering occupant behaviour

Energy consumption for space heating and cooling typically accounts for more than 40% of residential household energy consumption. An accurate and fast prediction of space heating and cooling loads aids energy conservation and carbon emission reduction by relieving the simulation burden for optimisation design, which consider various building characteristics combinations. This study aims to develop machine learning based load prediction model for residential building, five machine-learning models have been utilised for the prediction of residential building space heating and cooling load intensities, with occupant behaviour innovatively accounted as predictor variable. Their prediction performances are compared with each other. The five machine-learning models used in this study are linear kernel support vector regression, polynomial kernel support vector regression, Gaussian radial basis function kernel support vector regression, linear regression, and artificial neural networks. The results indicate that the Gaussian radial basis function kernel support vector regression is the best-performing model, with training time of less than 35s as well as less than 4% normalised mean absolute error and normalised root-mean-square error for both cooling and heating load prediction. The sample size of training and validation set for Gaussian radial basis function kernel support vector regression model is suggested as 200 samples. A data-driven machine-learning-based prediction model is an alternative to complex simulation tools in aiding the decision making of both building design and retrofit processes.

Quantitative Assessment of Speech in Cerebellar Ataxia Using Magnitude and Phase Based Cepstrum.

The clinical assessment of speech abnormalities in Cerebellar Ataxia (CA) is time-consuming and inconsistent. We have developed an automated objective system to quantify CA severity and thereby facilitate remote monitoring and optimisation of therapeutic interventions. A quantitative acoustic assessment could prove to be a viable biomarker for this purpose. Our study explores the use of phase-based cepstral features extracted from the modified group delay function as a complement to the features obtained from the magnitude cepstrum. We selected a combination of 15 acoustic measurements using RELIEF feature selection algorithm during the feature optimisation process. These features were used to segregate ataxic speakers from normal speakers (controls) and objectively assess them based on their severity. The effectiveness of our study has been experimentally evaluated through a clinical study involving 42 patients diagnosed with CA and 23 age-matched controls. A radial basis function kernel based support vector machine (SVM) classifier achieved a classification accuracy of 84.6% in CA-Control discrimination [area under the ROC curve (AUC) of 0.97] and 74% in the modified 3-level CA severity estimation (AUC of 0.90) deduced from the clinical ratings. The strong classification ability of selected features and the SVM model supports this scheme's suitability for monitoring CA related speech motor abnormalities.

Adaptive data-driven nonlinear synchro squeezed transform with single class radial basis function kernel support vector machine applied to wind turbine planetary gearbox fault diagnostics

Planetary stage gears operated at low rotational speed and varying wind speed result variation in load. Variable speed and variable load induce nonstationary operating conditions. Vibration signal measured from Wind power gear transmission systems are embedded with multiple sources of vibration and attenuated considerably as it travels from source of vibration to measuring point. Efficacious multi-component decomposition without mode mixing ensures the accurate fault signature recognition. Synchro squeezing transform is the promising tool that represents the ridges with high resolution in time as well as in frequency axis. An efficient vibration analysis technique, short windowed Fourier synchro squeezing transform with nonlinear radial basis function kernel support vector machine is proposed to detect the mechanical faults in low speed planetary stage of wind turbines. Raw vibration is modeled in time–frequency plane to extract fault pattern signatures effectively with high resolution by adapting an empirical nonlinear tool synchro squeezing transforms. Amplitude modulation and frequency modulation parameters are sculpted from instantaneous amplitude and instantaneous phase, frequency. Hybrid feature space with signal attributes, statistical moments, and randomness measures are extricated from amplitude modulation-frequency modulation components. Single class radial basis function support vector machine is trained with hybrid features. The fault detection accuracy of the proposed method is compared with the standard variants of empirical mode decomposition. The proposed short windowed Fourier synchro squeezing transform-radial basis function kernel support vector machine shows 98.2% accuracy, 98% sensitivity, and 98% specificity.

Efficient operating room planning using an ensemble learning approach to predict surgery cancellations

Cancelled surgeries cause insufficient use of operating rooms, surgeons, nurses, equipment and other hospital resources. Operating rooms are one of the most expensive resources in any healthcare system. The average amount of cost for any operating room for a less complex procedure is around $29/min, while the more complex procedures cost up to $80/min. Many studies have been conducted to analyze and understand the reasons behind surgery cancellation; however, a handful of studies aimed to predict which patients have a large risk of cancellation. We used four different traditional data mining techniques—Conditional Inference Tree, C5.0, logistic regression, and Radial Basis Function Kernel Support Vector Machine—to predict surgical cancellations to create efficient surgical patients’ schedules. Then, a stacking generalization ensemble machine was developed and compared to the traditional methods. Three different scheduling scenarios were developed and tested using discrete event simulation (DES). The stacking generalization ensemble machine outperformed all traditional data mining techniques and the benchmark algorithm with an accuracy of 95% and area under the curve (AUC) of 96%. The proposed simulation-based optimization scheduling technique increased the operating room use by 13%. Surgeries cancellation prediction leads to efficient scheduling that ultimately leads to reductions in expenditure.Highlights:Created a robust framework to increase OR utilizationPredicted surgery cancellation using data mining approachesCreated a novel stacking ensemble machineEnhanced the prediction power through ensemble learning that outperformed the other algorithmsDetermined overbooking opportunities through Simulation-based optimization

Prognosis of cervical myelopathy based on diffusion tensor imaging with artificial intelligence methods.

Diffusion tensor imaging (DTI) has been proposed for the prognosis of cervical myelopathy (CM), but the manual analysis of DTI features is complicated and time consuming. This study evaluated the potential of artificial intelligence (AI) methods in the analysis of DTI for the prognosis of CM. Seventy-five patients who underwent surgical treatment for CM were recruited for DTI imaging and were divided into two groups based on their one-year follow-up recovery. The DTI features of fractional anisotropy, axial diffusivity, radial diffusivity, and mean diffusivity were extracted from DTI maps of all cervical levels. Conventional AI models using logistic regression (LR), k-nearest neighbors (KNN), and a radial basis function kernel support vector machine (RBF-SVM) were built using these DTI features. In addition, a deep learning model was applied to the DTI maps. Their performances were compared using 50 repeated 10-fold cross-validations. The accuracy of the classifications reached 74.2%±1.6% for LR, 85.6%±1.4% for KNN, 89.7%±1.6% for RBF-SVM, and 59.2%±3.8% for the deep leaning model. The RBF-SVM algorithm achieved the best accuracy, with sensitivity and specificity of 85.0%±3.4% and 92.4%±1.9% respectively. This finding indicates that AI methods are feasible and effective for DTI analysis for the prognosis of CM.

Predictive modelling and analytics for diabetes using a machine learning approach

Diabetes is a major metabolic disorder which can affect entire body system adversely. Undiagnosed diabetes can increase the risk of cardiac stroke, diabetic nephropathy and other disorders. All over the world millions of people are affected by this disease. Early detection of diabetes is very important to maintain a healthy life. This disease is a reason of global concern as the cases of diabetes are rising rapidly. Machine learning (ML) is a computational method for automatic learning from experience and improves the performance to make more accurate predictions. In the current research we have utilized machine learning technique in Pima Indian diabetes dataset to develop trends and detect patterns with risk factors using R data manipulation tool. To classify the patients into diabetic and non-diabetic we have developed and analyzed five different predictive models using R data manipulation tool. For this purpose we used supervised machine learning algorithms namely linear kernel support vector machine (SVM-linear), radial basis function (RBF) kernel support vector machine, k-nearest neighbour (k-NN), artificial neural network (ANN) and multifactor dimensionality reduction (MDR).

Dew point pressure prediction based on mixed-kernels-function support vector machine in gas-condensate reservoir

Dew point pressure, at which the first condensate liquid comes out of solution in gas condensate reservoir, is a vital parameter for fluid characterization, field development, reservoir management and facility construction. Fast and accurate measurement of dew point pressure is always a challenge. Laboratory measurement can give accurate dew point pressure, but are expensive and time consuming. Equation of state is an alternative way, but can not converge in light oil and gas condensate reservoirs. Different empirical correlations have been built up between reservoir properties, fluid composition and dew point pressure. However, those correlations do not accurately reflect complex, non-linear relationships between them. With the development and improvement of artificial neural networks, different neural networks; such as multilayer perceptron neural network, radial basis function neural network, and gene expression programming can be used to describe complex relationships. Recently, one popular machine learning algorithm-(support vector machine) attracts attention due to its strong generalization ability. In this paper, we introduce a mixed kernel function based support vector machine (MKF-SVM), which has both strong interpolation and extrapolation abilities. This support vector machine model was trained and tested using 564 measurements of dew point pressure.The performance of this model is compared against four well known empirical correlations for dew point pressure calculation. The result, high R2=0.9150, low root mean square error RMSE = 476.392 and low average absolute percent relative error (AAPE = 7.01%) indicates good performance of mixed kernel function based support vector machine (MKF-SVM).

Recognition of Sign Language Alphabets and Numbers based on Hand Kinematics using A Data Glove

This paper reports real-time recognition of Indian and American sign language alphabets and numbers based on hand kinematics assessment. The finger and wrist joint angles were acquired using an indigenously developed data glove. The data set was for single handed Indian sign language alphabets (C, I, J, L, O, U, Y, W), American sign language alphabets (A to Z) and sign numbers (0 to 9). The data were pre-processed through a moving average filter and standardized feature scaling methods. The glove was able to measure the finger joint angles with an accuracy±standard deviation for metacarpophalangeal (MCP) joint±2.14°, proximal inter phalangeal (PIP) joint± 1.73° and distal inter phalangeal (DIP) joint± 1.49°, during flexion/extension and abduction/adduction movements. A radial basis function kernel support vector machine with 10-fold cross validation was used for recognition. An average recognition rate of 96.7% was achieved. Using a label matching and speech data base, the recognized alphabets and numbers were translated into speech.

A personalized point-of-interest recommendation model via fusion of geo-social information

Recently, as location-based social networks (LBSNs) rapidly grow, general users utilize point-of-interest recommender systems to discover attractive locations. Most existing POI recommendation algorithms always employ the check-in data and rich contextual information (e.g., geographical information and users’ social network information) of users to learn their preference on POIs. Unfortunately, these studies generally suffer from two major limitations: (1) when modeling geographical influence, users’ personalized behavior differences are ignored; (2) when modeling the users’ social influence, the implicit social influence is seldom exploited. In this paper, we propose a novel POI recommendation approach called GeoEISo. GeoEISo achieves three key goals in this work. (1) We develop a kernel estimation method with a self-adaptive kernel bandwidth to model the geographical influence between POIs. (2) We use the Gaussian radial basis kernel function based support vector regression (SVR) model to predict explicit trust values between users, and then devise a novel trust-based recommendation model to simultaneously incorporate both the explicit and implicit social trust information into the process of POI recommendation. (3) We develop a unified geo-social framework which combines users’ preference on a POI with the geographical influence as well as social correlations. Experimental results on two real-world datasets collected from Foursquare show that GeoEISo provides significantly superior performances compared to other state-of-the-art POI recommendation models.

Object-Based Land-Cover Supervised Classification for Very-High-Resolution UAV Images Using Stacked Denoising Autoencoders

Over the last decade, object-based image classification (OBIC) has become a mainstream method in remote sensing land-use/land-cover applications. Many supervised classification methods have been proposed in the OBIC framework. However, most did not use deep learning methods. In this paper, a new deep-learning-based OBIC framework is introduced. First, we segment the original image into objects by graph-based minimal-spanning-tree segmentation algorithm. Second, we extract the spectral, spatial, and texture features for each object. Then we put all features into stacked autoencoders (SAE) or stacked denoising autoencoders (SDAE) network, and trained the parameters of the network using training samples. Finally, all objects were classified by the network. Based on our SAE/SDAE OBIC framework, we achieved 97% overall accuracy when classifying an UAV image into five categories. In addition, our experiment shows that our framework increases overall accuracy by approximately 6% when compared to the linear support vector machine (linear SVM) and radial basis function kernel support vector machine (RBF SVM) algorithms when sufficient training samples are lacking.