The data mining techniques such as Artificial Neural network algorithm (ANN) and support vector machine (SVM) based models are utilized in this study. Three different kernel functions were used as SVM kernel functions which are polynomial, Radial Basis Function (RBF), and Pearson VII function-based Universal Kernel (PUK). The SVM models based on these three kernel functions and ANN model were used in forecasting the cotton yarn properties from cotton fiber properties. These cotton fiber properties and their corresponding yarn properties were collected from two different spinning mills in Zhengzhou, and Xuchang, in Henan Province, China, and the third dataset was obtained from the published paper. These cotton fiber properties were measured by HVI and AFIS instruments and were fiber length UHML (mm), length uniformity (%), Short Fiber Content SFC (%), micronaire (M), fiber strength (cN/dtex), elongation (%), yellowness (+b), and reflectance (Rd), trash content (Cnt), and Neps for HVI and were fiber length UHML (mm), Short Fiber Content SFC (%), maturity, fineness, neps, seed coat neps and trash for AFIS. The yarn properties that were used as output or targets of prediction were yarn unevenness, hairiness, the thin places, the thick places, the neps, yarn tenacity (cN/tex), the Strength CV%and yarn elongation. The results of the prediction of cotton carded yarn properties such as yarn unevenness, hairiness, the thin places, the thick places, the neps, yarn tenacity (cN/tex), as well as the Strength CV%indicated that the SVM model based on PUK were better than SVM model based on RBF. Both SVM models based on RBF and PUK gave a better predictive performance than an ANN model. The prediction results by using SVM based on polynomial was worse in comparison with the two SVM model based on RBF and PUK as well as ANN model. Similar findings were obtained when the SVM model based on RBF, PUK and polynomial kernels as well as ANN model were used to predict the ring and compact yarn properties such as yarn unevenness, hairiness, yarn tenacity (cN/tex) and yarn elongation. The predicted results show the SVM model based on RBF and PUK are better than an ANN model, while the SVM model based on polynomial showed the worst model results. As to find more accurate results between the comparison of the SVM model based on RBF, PUK and polynomial kernel as well as ANN model, the attempt was made to predict the yarn properties from cotton fiber properties collected from one published paper. The results also indicated that the SVM model based on RBF and PUK was better than an ANN model. The SVM model based on polynomial yielded the worst results of the models. However, the general comparison of the results indicates that the SVM models based on RBF and PUK Performs yarn properties forecasts more accurately than ANN. In order to know the relationship between fiber and yarn properties and find which fiber properties are the actual factors that affect of yarn properties, two different types of feature selection were used; feature ranking techniques and the Genetic Algorithm (GA) for feature selection.In order to know the ranking performance of the Feature ranking techniques, four different based feature ranking algorithms such as principal component analysis (PCA) attribute ranking, SVM weight-based ranking, Relief, and Correlation-based feature selection were used and compared with each other. However, the results indicated that there was obvious difference from each ranking method in performing the ranking of data and find the relationship between fiber and yarn properties. The general performance indicated that all four ranking methods can perform the ranking and find the relation between the fiber and yarn properties well. On the other side, a proposed hybrid approach of genetic algorithm for feature selection (GA) combined with support vector machines for regression (SVMR) was applied in this work to optimize the dataset of fiber properties and predict the yarn tenacity property. This hybrid approach was compared with a noisy model of SVMR that used all dataset of fiber properties as input in prediction. The GA for feature selection was used as preprocessing stage that aims to find and select the best attributes or variables that most effect or related to prediction of yarn tenacity. The prediction results of the hybrid approach of GA and SVMR were0.449for RMSE,0.420for MSE,3.45%for RE,0.226for SE and0.799for R. The prediction results of the noisy SVMR model were0.527for RMSE,0.474for MSE,3.73%for RE,0.329for SE and0.800for R. However, from these results, the hybrid approach showed better predictive performance than the noisy model. The results, nevertheless, indicated the suitability of GA for feature selection in the selection of the best fiber properties attributes that gives as the preferred performance and high accuracy in prediction of yarn tenacity.