Abstract:
There is an effect from the apparel industry to the total export earnings of Sri Lanka. This
paper consists of a case study relevant to the product quality of a leading apparel
manufacturing plant in Sri Lanka. Quality is measured using daily first time through
(FTT) percentage that calculated using daily output and number of defects. The main
purpose of this study is to identify the factors affecting the daily FTT and to build a model
to forecast daily FTT status. Factors affecting daily FTT were identified using multiple
linear regression and Yeo Johnson power transformation methods. According to the
attained results material defect, incorrect fabric direction, missing trim, and needle cut
were identified as the influential factors for daily FTT. The factory standard is to maintain
FTT 98% or above. For lower FTT measurement out and color, shading was affected.
Data mining techniques were applied due to the violation of statistical assumptions in the
aforementioned traditional methods. Classification tree and Probabilistic neural network
(PNN) techniques were applied to the classes of daily FTT values of high and low as a
classification problem based on the factory standard level. The under-sampling technique
was used due to a class imbalance problem. The best split attribute was the number of
damages and daily output was the afterward split attribute in the classification tree. In
PNN the best model was selected using adjustment of the spread parameter from 0 to 1.
Least false positive and false negative values were in the spread value 0.80 with the
highest true positive and true negative values. PNN model consists of 1857 and 2 hidden
neurons in the first and second hidden layers respectively. Accuracy was 0.98 in the
classification tree which is higher than the accuracy of PNN, which of 0.93. However,
both models can be used in forecasting with high accuracy. This research can benefit the
apparel field to get remedial actions before arising quality issues.
Keywords: Apparel industry, First time through, Data mining, Classification tree,
Probabilistic neural network
