PREDICTING YARN QUALITY PROPERTIES VIA OVERCOMING THE MULTICOLLINEARITY OF COTTON FIBER PROPERTIES

When predicting yarn quality properties, the collinearity or common variance of cotton fiber characteristics as predictors, result in unreal regression models. The approach of this study is using Principle Component Analysis (PCA) to avoid this issue by extracting independent factors in their effect from each other summarizes cotton fiber properties. Four lint cotton grades of five of Egyptian cotton varieties belong to Extra-long (Giza 88 and Giza 92) and Long staple (Giza 86, Giza 90 and Giza 95) classes used to perform fiber tests. Cotton Classification System (CCS-V5.3) used to measure cotton fiber characteristics as predictors. Yarn strength in terms of Lea product, single yarn strength and yarn unevenness of Ne 40 and 60 counts of ring spun yarns were the dependent variables. The results showed significant intercorrelations matrix among CCS measurements. The initial solution extracted only three factors that have eigenvalues more than 1.00. These 3 factors accounted for 89.716 % of the common variance shared by all measurements. The communalities or % variance in each cotton fiber measurement of CCS accounted for by the three factors was not the same. The 3 factors as predictors could predict yarn quality characteristics significantly, and with high contributions (% R 2 ). But % R 2 valued less than that of ordinary regression models. This audit is a satisfactory improvement to predict yarn quality characteristics from cotton fiber properties accurately.