Estimation of parameters affecting washing performance of different stain groups in washing machines with supervised learning algorithms

The washing performance of household washing machines must comply with certain limit values, such eco-design requirements in Europe. The limit values determined for the washing performance of stains in washing machines must be obtained according to the EN 60456 standard. Washing is carried out at various levels of factors affecting washing performance, such as washing time, water level, detergent amount, motor intensity, amount of washed load per volume and temperature. In this thesis study, washing performance values obtained at quarter load, half load and full load on fabrics with sebum, carbon, blood, cocoa and red wine stains were obtained from Vestel company. The washing performance value for 5 stains and the sum of the washing performance value of all stains were obtained, and a total of 18 data sets were obtained for 3 separate loads. The resulting data sets were modeled with multiple linear regression, regression tree, bagging, random forest regression, XGBoost regression, support vector regression (linear kernel and polynomial kernel) and k-nearest neighbor regression algorithms. To compare the performances of all models, mean square error, root mean square error and mean absolute error metrics were used. Out of 18 datasets, random forests regression was found to be the best model in 8, bagging in 7 and XGBoost in 3. As a result, tree-based algorithms have come to the fore as the best models.