| Summary: | The availability of Big Data (BD) is a critical factor that semiconductor makers can leverage to increase their forecast accuracy. Manufacturers can use this data to better forecast future consumer demand in semiconductor manufacturing and make best guesses about the quantities of each product variant to create via selective manufacturing based on calculated risk. The potential of five distinct machine learning models was investigated in this study: Linear Model Regression, Multi-layer Perceptron Regressor Model (MLP), K-Nearest Neighbour Regressor Model (KNN), Hoeffding Tree Regressor Model (HT), and Hoeffding Adaptive Tree Regressor (HAT). Three statistical metrics were used to assess the accuracy of constructed models: mean absolute error (MAE), root mean squared error (RMSE), and residual distribution. Meta, Imblearn, Expert, and Ensemble were employed in this experiment to enhance the machine learning technique. Along with the model, a drift detection method has been implemented such as ADWIN into our experiments to observe rapid changes in the generated BD. The comparison of research showed that the Linear Model Regression model using the Box-Cox improvement method outperformed MLP, KNN, HT, and HAT. In this study, the relationship model has been developed using the incremental data mining River's library in open source software Python. The overall results indicated that the Linear Model Regression model (Box-Cox approached) model could be successfully applied in the semiconductor supply chain by using the generated BD.
|
|---|