The challenges of defects in manufacturing and assembly processes in optoelectronic industry continue to persist. Defective products cause increased time to completion (cycle time), energy consumption, cost, and loss of precious material. A complex laser assembly process is studied with the aim of minimising the generation of defective laser modules. Subsequently, relevant data were gathered to investigate machine learning and artificial intelligence methods to predict the output beam power of the module during the assembly process. The assembly process was divided into a number of chain steps, where we implemented a bespoke framework of hybrid feature selection method alongside artificial neural networks (ANNs) to formulate the statistical inferences. A review of existing learning methods in manufacturing and assembly processes enabled us to select XGBoost and random forest regression (RFR) as the two methods to be compared with ANN, based on their capabilities; ANN outperformed both of them, as it avoided overfitting and scored similar test metrics in the majority of the assembly steps. The results of the proposed solution have been validated in a real production dataset, even showing good predictive capability in the early steps of the assembly process where the available information is limited. Furthermore, the transferability of the framework was validated by applying the proposed framework to another product that follows a similar assembly process. The results indicated that the proposed framework has the potential to serve as the foundation for further research on laser modules’ sophisticated and multi-step assembly lines.