Feasibility analysis of dye color modeling in clothing industries with machine learning models

The latest available expert systems used for the dyeing process yields only about 50% accuracy in choosing dye codes from multiple pigments given a targeted sample on cloth. Hence, leading to anomalies in dyeing and ultimately thousands of dollars are lost in the daily process. To counter this state of affairs, we are proposing two methods. Firstly, the expert system is unchanged, and the auxiliary reliability analysis system is added to reduce the wear and collect a large amount of data. Secondly, we establish several machine learning models and the colorist (i.e., a combination of dye codes) will get from the dyeing formula in the model. In our evaluation, we try to predict the dye color combinations using both statistical and neural network approaches. The best result was achieved by the SVM approach in which the system gets a 7% error rate in prediction of the combination of dye codes from multiple pigments. This indicates a strong potential of our approach in this application.