Bridge Management Systems (BMSs) are a common tool for bridge management to extend the life cycle of bridge networks. However, the reliability of current BMS outcomes is doubtful. This is because: (1) Overall Condition Rating (OCR) method cannot represent individual bridge elements' condition and is unable to represent condition ratings of bridge elements in lower Condition States and due to (2) insufficient historical bridge records available. A long-term Performance Bridge (LTPB), i.e. deterioration, model is the most crucial component and decides level of reliability of long-term bridge needs. Recent development of an AI-based bridge deterioration model was undertaken to minimise these shortcomings. However, this model is computationally costly due to the process of Neural Network, generating a large data output. To improve the neural network process, optimisation is required. The hybrid optimisation method is proposed in this paper to filter out feasible condition ratings as input for long-term prediction modelling.