Genetic Algorithm Optimized Support Vector Research Articles

Hybrid machine learning-based model for predicting chloride ion concentration in coral aggregate concrete and its ethically aligned graphical user interface design

The chloride ion concentration distribution in coral aggregate concrete (CAC-CC) holds significant importance for evaluating CAC performance, assessing rebar corrosion, and understanding chloride ion diffusion. Thus, this study employs hybrid machine learning algorithms to establish predictive models for CAC-CC under three environmental conditions (salt spray zone, tidal zone, and underwater zone) and evaluates the accuracy and generalization ability of these models. The results demonstrate that the genetic algorithm-optimized support vector regression (GA-SVR) model provides predictions closer to the actual values, exhibiting smoother error fluctuations. The GA-SVR model also exhibits smaller mean and standard deviation in the residual distribution. Furthermore, the GA-SVR model excels in five performance evaluation metrics, with R2, MAE, MAPE, MSE, and RMSE values of 0.986, 1.25e-2, 3.24e-2, 0.276e-3, and 1.66e-2, respectively. Hence, the GA-SVR model emerges as the optimal choice for CAC-CC prediction. Additionally, this work utilizes Shapley Additive Explanations (SHAP) to assess the contribution of eight features. The analysis reveals that the water-binder ratio and the pre-wetted water to total water ratio are the two most critical features. Moreover, an increase in the pre-wetted water to total water ratio and a decrease in the water-cement ratio hinder chloride ion diffusion. Mechanistic analysis of the feature contribution to CAC-CC is further explored in this study. Based on the GA-SVR model, a graphical user interface for CAC-CC has been developed, enabling visualization of CAC-CC predictions. This research introduces a novel approach to optimize CAC performance and predict chloride ion concentration, laying the foundation for CAC application in reef construction.

A fault diagnosis method for building electrical systems based on the combination of variational modal decomposition and new mutual dimensionless

The fault diagnosis of building electrical systems are of great significance to the safe and stable operation of modern intelligent buildings. In this paper, it has many problems, such as various fault types, inconspicuous fault characteristics, uncertainty of fault type and mode, irregularity, unstable signal, large gap between fault data classes, small gap between classes and nonlinearity, etc. A method of building electrical system fault diagnosis based on the combination of variational mode decomposition and mutual dimensionless indictor (VMD-MDI) and quantum genetic algorithm-support vector machine (QGA-SVM) is proposed. Firstly, the method decomposes the original signal through variational modal decomposition to obtain the optimal number of Intrinsic Mode Function(IMF) containing fault feature information. Secondly, extracts the mutual dimensionless indicator for each IMF. Thirdly, the optimal penalty coefficient C of the support vector machine and the parameter gamma (gamma) in the radial basis kernel function are selected by the quantum genetic algorithm. Finally, SVM optimized by the QGA is used to identify and classify the faults. By applying the proposed method to the experimental platform data of building electrical system, and compared with the traditional feature extraction method Empirical Mode Decomposition (EMD), Singular Value Decomposition(SVD), Local Mean Decomposition(LMD). And compared with traditional SVM, Genetic Algorithm optimized Support Vector Machine (GA-SVM), One-Dimensional Convolutional Neural Network (1DCNN) for fault classification methods. The experimental results show that the method has better effect and higher accuracy in fault diagnosis and classification of building electrical system. Its average test accuracy can reach 91.67%.

Disaster Risk Regionalization and Prediction of Corn Thrips Combined with Cloud Model: A Case Study of Shandong Province, China

Corn thrips do serious harm to the yield and quality of corn. In this paper, the Shandong Province of China was taken as the study area. Based on the data of the occurrence of corn thrips in Shandong Province, a risk regionalization model was established by using eight indicators under four categories of hazard, sensitivity, vulnerability and the disaster prevention and mitigation capacity of diseases and pests on a monthly time scale. Firstly, the cloud model was introduced to determine the weight of each indicator, and then the risk regionalization of the corn thrips disaster in Shandong Province was carried out using the weighted percentage method, the weighted comprehensive evaluation method and the natural disaster risk index method. Finally, combined with the collected data, the disaster prediction of corn thrip occurrence degree was realized based on multiple linear regression, genetic algorithm optimized back-propagation neural network and genetic algorithm optimized support vector machine methods. The results show that: (1) the risk of Corn thrips disaster is mainly concentrated in the central and western parts of Shandong Province. Heze City is a high-risk area. Liaocheng City, Dezhou City, Jinan City and Weifang City are relatively high-risk areas. (2) By comparing the prediction accuracy of the three models, it was determined that the genetic algorithm optimized support vector machine model has the best effect, with an average accuracy of 79.984%, which is 7.013% and 22.745% higher than that of the multiple linear regression and genetic algorithm optimized back-propagation neural network methods, respectively. The results of this study can provide a scientific basis for fine prevention of corn thrips in Shandong Province.

Rapid determination of Panax notoginseng origin by terahertz spectroscopy combined with the machine learning method

Panax notoginseng is a valuable herb with geographical indication, and the quality and price of P. notoginseng from different origins are very different. Therefore, this paper proposes a rapid and accurate method for identifying the origins of P. notoginseng by collecting the roots of P. notoginseng. This paper improves the whale optimization algorithm in terms of global convergence and convergence speed, introduces the Levy flight strategy and reconstructed whale synergy factor A, and applies it to the parameter optimization of support vector machines, to obtain a high-performance classification model. The improved whale optimization algorithm model identifies the origin of P. notoginseng by discriminating their terahertz spectra. Compared with the commonly used genetic algorithm and the original whale optimization algorithm, improvement in the whale optimization algorithm was able to avoid falling into local optimum solutions more effectively while having a high convergence rate. Accordingly, the improved whale optimization algorithm optimized support vector machine model obtained an overall accuracy of 98.44%, which was significantly higher than the 95.31% overall accuracy of the genetic algorithm optimized support vector machine model and the 96.88% overall accuracy of the whale optimization algorithm optimized support vector machine model. It was concluded that terahertz spectroscopy together with machine learning would be a promising technique for identifying the origins of P. notoginseng.

Concentration Prediction of Polymer Insulation Aging Indicator-Alcohols in Oil Based on Genetic Algorithm-Optimized Support Vector Machines.

The predictive model of aging indicator based on intelligent algorithms has become an auxiliary method for the aging condition of transformer polymer insulation. However, most of the current research on the concentration prediction of aging products focuses on dissolved gases in oil, and the concentration prediction of alcohols in oil is ignored. As new types of aging indicators, alcohols (methanol, ethanol) are becoming prevalent in the aging evaluation of transformer polymer insulation. To address this, this study proposes a prediction model for the concentration of alcohols based on a genetic-algorithm-optimized support vector machine (GA-SVM). Firstly, accelerated thermal aging experiments on oil-paper insulation are conducted, and the concentration of alcohols is measured. Then, the data of the past 4 days of aging are used as the input feature of SVM, and the GA algorithm is utilized to optimize the kernel function parameter and penalty factor of SVM. Moreover, the concentrations of methanol and ethanol are predicted, after which the prediction accuracy of other algorithms and GA-SVM are compared. Finally, an industrial software program for predicting the concentration of methanol and ethanol is established. The results show that the mean square errors (MSE) of methanol and ethanol concentration predictions of the model proposed in this paper are 0.008 and 0.003, respectively. The prediction model proposed in this paper can track changes in methanol and ethanol concentrations well, providing a theoretical basis for the field of alcohol concentration prediction in transformer oil.

Genetic algorithm–optimized support vector machine for real-time activity recognition in health smart home

Health smart home, as a typical application of Internet of things, provides a new solution for remote medical treatment. It can effectively relieve pressure from shortage of medical resources caused by aging population and help elderly people live at home more independently and safely. Activity recognition is the core of health smart home. This technology aims to recognize the activity patterns of users from a series of observations on the user’ actions and the environmental conditions, so as to avoid distress situations as much as possible. However, most of the existing researches focus on offline activity recognition, but not good at online real-time activity recognition. Besides, the feature representation techniques used for offline activity recognition are generally not suitable for online scenarios. In this article, the authors propose a real-time online activity recognition approach based on the genetic algorithm–optimized support vector machine classifier. In order to support online real-time activity recognition, a new sliding window-based feature representation technique enhanced by mutual information between sensors is devised. In addition, the genetic algorithm is used to automatically select optimal hyperparameters for the support vector machine model, thereby reducing the recognition inaccuracy caused by manual tuning of hyperparameters. Finally, a series of comprehensive experiments are conducted on freely available data sets to validate the effectiveness of the proposed approach.

3D Mineral Prospectivity Modeling for the Low-Sulfidation Epithermal Gold Deposit: A Case Study of the Axi Gold Deposit, Western Tianshan, NW China

The Axi low-sulfidation (LS) epithermal deposit in northwestern China is the result of geological controls on hydrothermal fluid flow through strike-slip faults. Such controls occur commonly in LS epithermal deposits worldwide, but unfortunately, these have not been quantitatively analyzed to determine their spatial relationships with gold distribution and further guide mineral prospecting. In this study, we conduct a 3D mineral prospectivity modeling approach for the Axi deposit involving 3D geological modeling, 3D spatial analysis, and prospectivity modeling. The spatial analysis of geometric features revealed the gold mineralization trends in convex segments (0–20 m) with a specific distance from fault 2, the lower interface of late volcanic phase, and the upper interface of phyllic alteration with steep slopes (>65°), implying that gold deposition was significantly controlled by the morphological characteristics and distance fields of geologic features. The present alteration–mineralization zone at Axi has a larger width in bending sites (sections No. 35–15 and No. 40–56) than elsewhere, indicating the location of two fluid conduits extending to depth. The prediction-area plots and receiver operating characteristic curves demonstrated that (genetic algorithm optimized support vector regression (GA-SVR)) outperformed multiple nonlinear regression and fuzzy weights-of-evidence, which was proposed as a robust method to solve complicated nonlinear and high-dimensional issues in prospectivity modeling. Our study manifests spatial controls of structure, host rock, and alteration on LS epithermal gold deposition, and highlights the capability of GA-SVR for identifying deposit-scale potential epithermal gold mineralization.

Remaining useful life prediction of lithium-ion batteries with adaptive unscented kalman filter and optimized support vector regression

To solve the problem of the inaccurate prediction on remaining useful life (RUL) for lithium-ion battery, we proposed an integrated algorithm which combines adaptive unscented kalman filter (AUKF) and genetic algorithm optimized support vector regression (GA-SVR). Firstly, the state space model with double exponential is established to describe the degradation of lithium battery. Then, the AUKF algorithm is introduced to update adaptively both the process noise covariance and the observation noise covariance. Next, the genetic algorithm is utilized to optimize the key parameters of SVR which realizes multi-step prediction. The effectiveness of the proposed method is verified by simulation experiments with NASA of battery dataset. Simulation results show that the proposed AUKF-GA-SVR achieves better prediction accuracy than existed methods such as unscented kalman filter, extended kalman filter, adaptive extended kalman filter (AEKF), adaptive unscented kalman filter, unscented kalman filter and relevance vector regression and AEKF-GA-SVR.

Research on GA-SVM Based Head-Motion Classification via Mechanomyography Feature Analysis.

This study investigated classification of six types of head motions using mechanomyography (MMG) signals. An unequal segmenting algorithm was adopted to segment the MMG signals generated by head motions. Three types of features (time domain, time-frequency domain and nonlinear dynamics) were extracted to construct five feature sets as candidate datasets for classification analysis. Genetic algorithm optimized support vector machine (GA-SVM) was used to classify the MMG signals. Three different kernel functions, different combinations of feature sets, different number of signal channels and training samples were selected for comparative analysis to evaluate the classification accuracy. Experimental results showed that the classifier had the best overall classification accuracy when using the radial basis function (RBF). Any combination of three different types of feature sets guaranteed an average accuracy of over 80%. In the case of the best combination (feature set 2 + 3 + 5), the classification accuracy was up to 88.2%. Using four channels to acquire MMG signal and no less than 60 training samples can assure a satisfactory classification accuracy.

Prediction Performance of Separate Collection of Packaging Waste Yields Using Genetic Algorithm Optimized Support Vector Machines

Understanding the drivers underlying waste production in general, and source-segregated waste in particular, is of utmost importance for waste managers. This work aims at evaluating the performance of support vector machines (SVM) models in the prediction of separate collection yields for packaging waste at municipal level. Two SVM models were developed for a case study of 42 municipalities simultaneously serviced by separate collection of packaging waste and by unsorted waste collection. The “SVM-fixed” model used a fixed set of 5 variables to predict collection yields, whereas the “SVM-optimal” model chose from a pool of 14 variables those that optimized performance, using a genetic algorithm. These SVM models were compared with 3 traditional regression models: the ordinary least square linear (OLS-L), the ordinary least square non-linear (OLS-NL) and robust regression. The robust regression model was further compared against the other regression models in order to assess the influence of the dataset outliers on the model performance. The coefficient of determination, R2, was used to evaluate the performance of these models. The highest performance was attained by the SVM-optimal model (R2 = 0.918), compared to the SVM-fixed model (R2 = 0.670). The performance of the SVM-optimal model was 42% higher than the best performing regression model, the OLS-NL model (R2 = 0.646). The differences in performance among the 3 regression models are small (circa 3%), whereas the exclusion of outliers improved their performance by 13%, indicating that outliers impacted more on performance than the type of traditional regression technique used. The results demonstrate that SVM model can be a viable alternative for prediction of separate collection of packaging waste yields and that there are nine important drivers that all together explain roughly 92% (R2 = 0.918) of the variability in the separate collection yields data.

Assessing the authenticity of black pepper using diffuse reflectance mid-infrared Fourier transform spectroscopy coupled with chemometrics

Diffuse reflectance mid-infrared Fourier transform spectroscopy (DRIFTS) coupled with chemometrics was used to examine the possibility of verifying the origin and authenticity of black pepper in this study. DRIFTS were obtained for 150 pure black pepper samples of different geographical origin (Kerala, HaiNan, DongNai, Sarawak, Fujian) and 1200 fake black pepper (adulterated with sorghum or Sichuan pepper, doping levels from 5% to 50%, respectively) by recording within a spectroscopic range of 4000–400 cm−1. Collected data were treated using principal component analysis (PCA), genetic algorithm optimized support vector machine (GA-SVM) and partial least squares discriminant analysis (PLS-DA). PCA model of the DRIFTS revealed the clustering trend of the pure and fake black pepper samples. Accurate and reliable classification results were obtained by DRIFTS combined GA-SVM and PLS-DA model. For the pure black pepper sample, the training set and prediction set of GA-SVM and PLS-DA models all achieved 100% correct recognition rate. For the fake black pepper sample, the total recognition rate of GA-SVM and PLS-DA model were greater than 98% for the training sets and more than 96% for the predicted sets, respectively. The results of this study demonstrated that DRIFTS method combined with chemometric tools can be successfully applied to detect mislabeling and adulteration of black pepper.

Estimation of low calorific value of blended coals based on support vector regression and sensitivity analysis in coal-fired power plants

Low calorific value (LCV) of blended coals is a vital property for the economy and efficiency of the coal-fired power plants. The experimental determination of the fuel calorific value is costly, while proximate analysis is easier and cheaper. Many linear correlations have been developed to describe the relationships between the calorific value and a few constituents of proximate and/or ultimate analysis, although some may be nonlinear. In this paper a specially designed support vector regression based approach (Sensitivity analysis weighted and genetic algorithm optimized support vector regression, SAW-GA-SVR) was applied to estimate the LCV based on proximate analysis of coal fed to a power plant. In this, nonlinear sensitivity coefficients were used as weighting factors and genetic algorithm was used for optimization. The prediction capability of the proposed model was compared with support vector regression (SVR), genetic algorithm optimized support vector regression (GA-SVR), genetic algorithm optimized back propagation neural network (GA-BPNN), and linear correlations from the published literature. According to the results obtained, the SAW-GA-SVR model has shown better capability of modeling and predicting the LCV of coals over the other methods.

Improvement of the regression model for spindle thermal elongation by a Boosting-based outliers detection approach

Data-driven regression methods are adept at modeling the spindle thermal elongation for error reduction purposes, but are limited in their generalization ability for the established model to take effect under different work conditions, where one prime cause is that the measurement errors and environmental disturbances would inevitably lead to successive outliers in the experimental training data, and the regression algorithms tend to over-fit to such outliers in the thermal datasets which are usually small. Here, an approach is proposed for the first time to detect and remove the suspected outliers in the spindle thermal data while preserve most of the informative data at the same time, so that the disturbances from the outliers can be minimized. The proposed approach for outliers detection is a combination of Boosting and SVR (BSOD), where the support vector machine for regression (SVR) is used as the weak learner in Boosting. Furthermore, the SVR, which is now widely considered as the most suitable regression tool for mapping the temperature-thermal elongation relationship for a spindle, is adopted to regression model the spindle thermal elongation, where the genetic algorithm (GA) is employed to determine the optimal parameter setting for the modeling. Effectiveness of the BSOD approach was verified using four experimental datasets from two different spindles of precision boring machines under different work conditions. The results showed that the BSOD approach reduced the mean squared error MSE of the genetic algorithm optimized support vector machine estimation for unseen thermal elongation datasets in two cases by 66.09% and 49.13%, respectively. This research provides a feasible way to enhance the modeling quality for the thermal elongation of spindles during working process.

Comparison of three officinal species of Callicarpa based on a biochemome profiling strategy with UHPLC-IT-MS and chemometrics analysis

Traditional Chinese medicine (TCM) materials with closely related species are frequently fungible in clinical use. Therefore, holistic comparison of the composition in bioactive compounds is essential to evaluate whether they are equivalent in efficacy. Taking three officinal species of Callicarpa as a case, we proposed and validated a standardized strategy for the discrimination of closely related TCM materials, which focused on the extraction, profiling and multivariate statistical analysis of their biochemome. Firstly, serial liquid-liquid extractions were utilized to prepare different batches of Callicarpa biochemome, and the preparation yields were utilized for the normalization of sampling quantity prior to UHPLC-IT-MS analysis. Secondly, 34 compounds, including 19 phenylethanoid glycosides, 10 flavonoids and 5 terpenoids, were identified based on an untargeted UHPLC-IT-MS method. Thirdly, method validation of linearity, precision and stability showed that the UHPLC-IT-MS system was qualified (R2>0.995, RSD<15%) for subsequent biochemome profiling. After PCA and PLS-DA analysis, 30 marker compounds were screened and demonstrated to be of good predictability using genetic algorithm optimized support vector machines. Finally, a heatmap visualization was employed for clarifying the distribution of marker compounds, which could be helpful to determine whether the three Callicarpa species are, in fact, equivalent substitutes. This study provides a standardized biochemome profiling strategy for systemic comparison analysis of closely related TCM materials, which shows promising perspectives in tracking the supply chain of pharmaceutical suppliers.

Identification of vehicle suspension shock absorber squeak and rattle noise based on wavelet packet transforms and a genetic algorithm-support vector machine

The squeak and rattle (S&R) noise of a vehicle’s suspension shock absorber substantially influences the psychological and physiological perception of passengers. In this paper, a state-of-the-art method, specifically, a genetic algorithm-optimized support vector machine (GA-SVM), which can select the most effective feature subsets and optimize the model’s free parameters, is proposed to identify this specific noise. A vehicular road test and a shock absorber rig test are conducted to investigate the relationship between these features, and then an approach for quantifying the shock absorber S&R noise is given. Pre-processed signals are decomposed through a wavelet packet transform (WPT), and two criteria, namely, the wavelet packet energy (WPE) and wavelet packet sample entropy (WPSE), are introduced as the feature extraction methods. Then, the two extracted feature sets are compared based on this genetic algorithm. Another advanced method, known as the genetic algorithm-optimized back propagation neural network (GA-BPNN), is introduced for comparison to illustrate the superiority of the newly developed GA-SVM model. The result shows that the WPSE can extract more useful features than the WPE and that the GA-SVM is more effective and efficient than the GA-BPNN. The proposed approach could be retrained and extended to address other fault identification problems.

Automatic epilepsy detection using wavelet-based nonlinear analysis and optimized SVM

Aiming at the problems of low accuracy, poor universality and functional singleness for seizure detection, an effective approach using wavelet-based non-linear analysis and genetic algorithm optimized support vector machine (GA-SVM) is proposed to deal with five challenging classification problems in this study. Instead of the traditional discrete wavelet transform (DWT), we attempt to explore the ability of double-density discrete wavelet transform (DD-DWT) to decompose the original EEG into specific sub-bands. The Hurst exponent (HE) and fuzzy entropy (FuzzyEn) are extracted as input features and then fed into two classifiers. On using these ranking non-linear features, the GA-SVM configured with fewer features is found to achieve the prominent classification performance for various combinations such as AB-CD-E, A-D-E, ABCD-E, C-E and D-E, achieving accuracies of 99.36%, 99.60%, 99.40%, 100% and 100%, respectively. The results have indicated that our scheme is not only appropriate in solving problems with multiple classes but also of lower complexity and better expansibility. These characteristics would make this method become an attractive alternative for actual clinical diagnosis.

Structural characterization and discrimination of Chinese medicinal materials with multiple botanical origins based on metabolite profiling and chemometrics analysis: Clematidis Radix et Rhizoma as a case study

Traditional Chinese medicines (TCMs)-based products are becoming more and more popular over the world. To ensure the safety and efficacy, authentication of Chinese medicinal materials has been an important issue, especially for that with multiple botanical origins (one-to-multiple). Taking Clematidis Radix et Rhizoma (CRR) as a case study, we herein developed an integrated platform based on metabolite profiling and chemometrics analysis to characterize, classify, and predict the “one-to-multiple” herbs. Firstly, the predominant constituents, triterpenoid saponins, in three Clematis CRR were rapid characterized by a novel UPLC-QTOF/MS-based strategy, and a total of 49 triterpenoid saponins were identified. Secondly, metabolite profiling was performed by UPLC-QTOF/MS, and 4623 variables were extracted and aligned as dataset. Thirdly, by using pattern recognition analysis, a clear separation of the three Clematis CRR was achieved as well as a total number of 28 variables were screened as the valuable variables for discrimination. By matching with identified saponins, these 28 variables were corresponding to 10 saponins which were identified as marker compounds. Fourthly, based on the relative intensity of the marker compounds-related variables, genetic algorithm optimized support vector machines (GA-SVM) was employed to predict the species of CRR samples. The obtained model showed excellent prediction performance with a prediction accuracy of 100%. Finally, a heatmap visualization was employed for clarifying the distribution of identified saponins, which could be useful for phytochemotaxonomy study of Clematis herbs. These results indicated that our proposed platform was a powerful tool for chemical profiling and discrimination of herbs with multiple botanical origins, providing promising perspectives in tracking the formulation processes of TCMs products.

Coke Rate Prediction Model Based on Genetic Algorithm Optimized Support Vector Machine

The accurate prediction model’s establishing of the blast furnace coke rate is important for optimizing the integrated production indicators of iron and steel enterprise. For the problem of accuracy of the model of coke rate, This paper established blast coke rate modeling with support vector machine algorithm, the model parameters of support vector machine was optimized by genetic algorithm, then a coke rate model based on support vector machine with the best parameters was built. Simulation results showed that: the forecasting model’s outcome, average absolute error and the mean relative error, was small which is based on genetic algorithm optimized SVM. coke rate model based on Genetic algorithm optimized support vector machine has high degree of accuracy and a certain practicality.

Characterization and classification of seven Citrus herbs by liquid chromatography–quadrupole time-of-flight mass spectrometry and genetic algorithm optimized support vector machines

Citrus herbs have been widely used in traditional medicine and cuisine in China and other countries since the ancient time. However, the authentication and quality control of Citrus herbs has always been a challenging task due to their similar morphological characteristics and the diversity of the multi-components existed in the complicated matrix. In the present investigation, we developed a novel strategy to characterize and classify seven Citrus herbs based on chromatographic analysis and chemometric methods. Firstly, the chemical constituents in seven Citrus herbs were globally characterized by liquid chromatography combined with quadrupole time-of-flight mass spectrometry (LC–QTOF-MS). Based on their retention time, UV spectra and MS fragmentation behavior, a total of 75 compounds were identified or tentatively characterized in these herbal medicines. Secondly, a segmental monitoring method based on LC-variable wavelength detection was developed for simultaneous quantification of ten marker compounds in these Citrus herbs. Thirdly, based on the contents of the ten analytes, genetic algorithm optimized support vector machines (GA-SVM) was employed to differentiate and classify the 64 samples covering these seven herbs. The obtained classifier showed good prediction performance and the overall prediction accuracy reached 96.88%. The proposed strategy is expected to provide new insight for authentication and quality control of traditional herbs.

Performance assessment of different data mining methods in statistical downscaling of daily precipitation

Summary In this paper, nonlinear Data-Mining (DM) methods have been used to extend the most cited statistical downscaling model, SDSM, for downscaling of daily precipitation. The proposed model is Nonlinear Data-Mining Downscaling Model (NDMDM). The four nonlinear and semi-nonlinear DM methods which are included in NDMDM model are cubic-order Multivariate Adaptive Regression Splines (MARS), Model Tree (MT), k -Nearest Neighbor ( k NN) and Genetic Algorithm-optimized Support Vector Machine (GA-SVM). The daily records of 12 rain gauge stations scattered in basins with various climates in Iran are used to compare the performance of NDMDM model with statistical downscaling method. Comparison between statistical downscaling and NDMDM results in the selected stations indicates that combination of MT and MARS methods can provide daily rain estimations with less mean absolute error and closer monthly standard deviation and skewness values to the historical records for both calibration and validation periods. The results of the future projections of precipitation in the selected rain gauge stations using A2 and B2 SRES scenarios show significant uncertainty of the NDMDM and statistical downscaling models.