Automating Labor Cost Classification for PCBA Manufacturing Using Machine Learning

Electronics manufacturing service providers must assign labor costs to every printed circuit board assembly (PCBA) quoted to a customer, this is often done by determining incremental costs for each component. Manual classification is slow and error-prone, requires significant expertise, inflates quote lead-time, and potentially misprices labor. This capstone builds a production ready machine learning pipeline that automates the task while remaining transparent to business users. 

Historical quoting records form the core data set. Text fields, categorical attributes, and numeric dimensions are cleaned, encoded, and re-sampled to correct class imbalance. A multi-input, 4-branch neural network embeds free text with custom tokenization, processes structured features with dense layers, and fuses all input streams to predict one of 37 labor cost codes for electronic components. The trained model, preprocessing, and post processing are all packaged as a single custom prediction routine and deployed on Google Vertex AI, delivering real time inference via REST or batch APIs that integrate directly into legacy quoting systems.