Manufacturing Lead Time Classification Using Support Vector Machine

Abstract

Product lead time is the indicator of manufacturing controllability, efficiency and performance in most manufacturing industry. The common lead time estimation mostly applied with knowledge-based approaches, but most of them have focused on supervised regression task to predict continuous lead time, which somehow only predicts the product lead time without identifying the critical lengthy manufacturing process. This paper aims to predict the manufacturing lead time of a product with supervised machine learning approach through classification of lead time in time range duration and able to identify the lengthy manufacturing product lead time. The machine learning model are developed with support vector machine algorithm using production and work order data as the model input. The target class is the total duration of product manufacturing time discretized into category attribute with different range of duration. The dataset extracted from work order and production data source and merge into a full dataset. Data preprocessing and feature engineering were applied to clean the dataset and boost the feature set to have 81 features in total. The final feature set was selected using recursive feature elimination approach feature selection technique. The final model applied support vector machine algorithm to perform the classification task and compared with Random Forest and Artificial Neural Network for performance. The final model scored 84.62% of weighted accuracy and able to predict the over-time class with over 65% of accuracy.