support vector regression (svr)

Support Vector Machines (SVMs) are a valuable tool in the food industry due to their ability to handle complex, nonlinear relationships between variables, even with limited datasets, high-dimensional data, and noisy data. This makes SVMs well-suited for applications such as food quality and safety assessment, sensory evaluation, process optimization, and food authentication. Accordingly, a new approach is introduced to predict different features of a traditional yogurt drink, also called doogh. The proposed model combines the principles of Support Vector Regression with fuzzy logic to handle uncertainty and approximate complex relationships between inputs (retentate, xanthan, and shelf-life) and target variables including viscosity, syneresis, color values, and total acceptability. The implemented approach is particularly useful when dealing with problems where the relationships are not easily captured by traditional mathematical models due to their non-linearity or imprecision. Also, it mitigates the limitations of data availability. The predictive ability of the proposed model has been evaluated in terms of MSE, R2, RMSE, and MAE when adding different noise levels. Additionally, the conditions necessary to attain optimized metric values have been found. At the optimum point, the viscosity, syneresis, L*, a*, b* and total acceptability are 19.70 mPa.s, 11.30%, 97.04, -1.43, 8.13, and 5.00, respectively. Besides, the findings indicate that samples containing 0.8% retentate, 0.4% xanthan, and a 31-day shelf-life exhibit the highest viscosity, while those with 0.6% retentate, 0.4% xanthan, and a 31-day shelf-life show the lowest syneresis. Moreover, samples with 0.7% retentate, 0.2% xanthan, and a 13-day shelf-life demonstrate the highest total acceptability.

The application of dynamic and continuous time series has drawn attention due to the complexities of the rainfall-runoff process and the simplification of multiple regression and static methods. On the other hand, forecasting river flow is one of the main topics in flood control. This paper reports the results of applying the SWAT hydrological model to analyze the rainfall-runoff relationship for a 6-year period in the Kahir catchment basin, Sistan and Baluchistan province. The model output was calibrated by the SUF12 optimizer algorithm, and the data entered the model again. Then, the model output was used to forecast future periods using Support Vector Regression (SVR) and essential codes in MATLAB. The acceptable results of the SVR model regarding data prediction can be used as another method to estimate parameters and inputs.

Stock markets have a key role in the economic situation of the countries. Thus one of the major methods of flourishing the economy can be getting people to invest their money in the stock market. For this purpose, reducing the risk of investment can persuade people to trust the market and invest. Hence, Productive tools for predicting the future of the stock market have an undeniable effect on investors and traders’ profit. In this research, a two-stage method has been introduced to predict the next week's index value of the market, and the Tehran Stock Exchange Market has been selected as a case study. In the first stage of the proposed method, a novel clustering method has been used to divide the data points of the training dataset into different groups and in the second phase for each cluster’s data, a hybrid regression method (HHO-SVR) has been trained to detect the patterns hidden in each group. For unknown samples, after determining their cluster, the corresponding trained regression model estimates the target value. In the hybrid regression method, HHO is hired to select the best feature subset and also to tune the parameters of SVR. The experimental results show the high accuracy of the proposed method in predicting the market index value of the next week. Also, the comparisons made with other metaheuristics indicate the superiority of HHO over other metaheuristics in solving such a hard and complex optimization problem. Using the historical information of the last 20 days, our method has achieved 99% accuracy in predicting the market index of the next 7 days while PSO, MVO, GSA, IPO, linear regression and fine-tuned SVR has achieved 67%, 98%, 38%, 4%, 5.6% and 98 % accuracy respectively. in this research we have tried to forecast the market index of the next m (from 1 to 7) days using the historical data of the past n (from 10 to 100) days. The experiments showed that increasing the number of days (n), used to create the dataset, will not necessarily improve the performance of the method.

Remote sensing image analysis can be carried out at the per-pixel (hard) and sub-pixel (soft) scales. The former refers to the purity of image pixels, while the latter refers to the mixed spectra resulting from all objects composing of the image pixels. The spectral unmixing methods have been developed to decompose mixed spectra. Data-driven unmixing algorithms utilize the reference data called training samples and end-members. The performance of algorithms using training samples can be negatively affected by the curse of dimensionality. This problem is usually observed in the hyperspectral image classification, especially when a low number of training samples, compared to the large number of spectral bands of hyperspectral data, are available. An unmixing method that is not highly impressed by the curse of dimensionality is a promising option. Among all the methods used, Support Vector Machine (SVM) is a more robust algorithm used to overcome this problem. In this work, our aim is to evaluate the capability of a regression mode of SVM, namely Support Vector Regression (SVR), for the sub-pixel classification of alteration zones. As a case study, the Hyperion data for the Sarcheshmeh, Darrehzar, and Sereidun districts is used. The main classification steps rely on 20 field samples taken from the Darrehzar area divided into 12 and 8 samples for training and validation, respectively. The accuracy of the sub-pixel maps obtained demonstrate that SVR can be successfully applied in the curse of dimensional conditions, where the size of the training samples (12) is very low compared to the number of spectral bands (165).

The tensile strength (TS) of rocks is an important parameter in the design of a variety of engineering structures such as the surface and underground mines, dam foundations, types of tunnels and excavations, and oil wells. In addition, the physical properties of a rock are intrinsic characteristics, which influence its mechanical behavior at a fundamental level. In this paper, a new approach combining the support vector regression (SVR) with a cultural algorithm (CA) is presented in order to predict TS of rocks from their physical properties. CA is used to determine the optimal value of the SVR controlling the parameters. A dataset including 29 data points was used in this study, in which 20 data points (70%) were considered for constructing the model and the remaining ones (9 data points) were used to evaluate the degree of accuracy and robustness. The results obtained show that the SVR optimized by the CA model can be successfully used to predict TS.

This research aims to describe a novel model, namely Hybrid Adaptive-Neuro Fuzzy Inference System-Particle Swarm Optimization (ANFIS-PSO), for predicting corrosion rate of 3C steel considering different marine environment factors. In the present research, five parameters (temperature, dissolved oxygen, salinity, pH, and oxidation–reduction potential) were used as input variables, with corrosion rate being the only output variable. In the proposed hybrid ANFIS-PSO model, the PSO served as a tool to automatically search for and update optimal parameters for the ANFIS, so as to improve generalizability of the model. Eeffectiveness of the hybrid model was then compared those to two other models, namely Adaptive-Neuro Fuzzy Inference System–Genetic Algorithm (ANFIS-GA) and Support Vector Regression (SVR) models, by evaluating their results against the same experimental data. The results showed that the proposed hybrid model tends to produce a lower prediction error than those of ANFIS-GA and SVR with the same training and testing datasets. Indeed, the hybrid ANFIS-PSO model provides engineers with an applicable and reliable tool to conduct real-time corrosion prediction of 3C steel considering different marine environment factors.

Displacements induced by earthquake can be very large and result in severe damage to earth and earth supported structures including embankment dams, road embankments, excavations and retaining walls. It is important, therefore, to be able to predict such displacements. In this paper, a new approach to prediction of earthquake induced displacements of slopes (EIDS) using hybrid support vector regression (SVR) with particle swarm optimization (PSO) is presented. The PSO is combined with the SVR for determining the optimal value of its user-defined parameters. The optimization implementation by the PSO significantly improves the generalization ability of the SVR. In this research, the input data for the EIDS prediction consist of values of geometrical and geotechnical input parameters. As an output, the model estimates the EIDS that can be modeled as a function approximation problem. A dataset that includes 45 data points was applied in current study, while 36 data points (80%) were used for constructing the model and the remainder data points (9 data points) were used for assessment of degree of accuracy and robustness. The results obtained show that the SVR-PSO model can be used successfully for prediction of the EIDS.

Prediction of river flow is one of the main issues in the field of water resources management. Because of the complexity of the rainfall-runoff process, data-driven methods have gained increased importance. In the current study, two newly developed models called Least Square Support Vector Regression (LSSVR) and Regression Tree (RT) are used. The LSSVR model is based on the constrained optimization method and applies structural risk minimization in order to yield a general optimized result. Also in the RT, data movement is based on laws discovered in the tree. Both models have been applied to the data in the Kashkan watershed. Variables include (a) recorded precipitation values in the Kashkan watershed stations, and (b) outlet discharge values of one and two previous days. Present discharge is considered as output of the two models. Following that, a sensitivity analysis has been carried out on the input features and less important features has been diminished so that both models have provided better prediction on the data. The final results of both models have been compared. It was found that the LSSVR model has better performance. Finally, the results present these models as a suitable models in river flow forecasting.

Toward green educational building development, windows are important design elements as the source of natural lighting and heating in classrooms. The amount of natural lighting and net heating received by a classroom in a year depends on the school location, weather conditions, as well as the window orientation and size. Schools in Iran consume a considerable amount of energy which is mostly supplied using nonrenewable fossil fuel resources. This energy consumption can be reduced through a well-designed daylighting approach. In this paper, in order to investigate the effects of window characteristics on construction and operational costs of schools, by varying the Window-to-Wall Ratio (WWR) and window orientation, 288 daylighting scenarios are generated for a typical standard classroom in a warm-dry climatic zone in central Iran. The DOE-2 software is utilized to estimate annual gas and electric consumption, for the generated scenarios over a period of 50 years. Considering the operation and construction cost, the best window facing and optimal range of WWR in each orientation is determined for the studied standard classroom. The results of simulated daylighting scenarios are then used to train regression based Support Vector Machines (SVMs) in order to show the feasibility of applying the Support Vector Regression (SVR) as an artificial intelligent system. The obtained results show that SVR as an architectural assistant performs well and the SVR-based predictor can rapidly, easily and accurately predict the operational and construction cost of a classroom just by determining the window size and installation face.