Skip to Main Content
Control chart patterns can be used to determine behavior of system. They are vital in process control as they are used in detecting the abnormalities which may occur. Accurate identification of these charts is necessary to the efficiency and reduction of system troubleshooting time. The accuracy of the classification depends largely on how noisy the signals in these charts are. If their noise ratio is very high, this suggests that reliable classification is almost impossible. One of the major difficulties lies in differentiation between increasing and decreasing patterns especially where gradients of inclination and declination are small. This paper describes an improvement in identifying highly noisy control chart patterns by utilizing features extraction in classification using neural networks in previous works. Features, which were founded useful for the classification, are mean, standard deviation, skewness, and kurtosis. The improvement can be summarized into two factors, the introduction of two more useful features, slope and Pearson correlation coefficient, and the additional transformation derived from the original signal. This work yields better performance than previous works which used the same data set by increasing the overall accuracy from 83.30% to 90.47%.