Cost prediction for product development using hybrid deep learning model: a meta-heuristic model

Accurate cost prediction at the component level is most critical for the assessment of economic viability, the estimation of cost rates at the component level, and the subsequent detection of opportunities for cost reduction during the development process. The traditional approaches are mostly heuristic and statistical, hence seldom accurate at early stages. To overcome those challenges, the paper introduced the core of the proposed approach the Hybrid Deep Learning model that combines the optimized Restricted Boltzmann Machines and Radial Basis Function Networks model proposed for accurate cost prediction. First and foremost, the methodology will preprocess the raw data by way of cleaning, reduction, and transformation stages. This is followed by feature extraction that includes LDA, central tendency, and information gain-based methods for identifying optimal features. The new improved Chi-square test is used thereafter in selecting only the relevant features from among those extracted. A voting classifier accumulates outputs from both models to enhance prediction accuracy. Furthermore, the Self-Improved Battle Royale Optimizer (SI-BRO) fine-tunes the performance of the RBM in terms of optimizing the effectiveness of the model. The results for the proposed model, implemented in MATLAB, are as follows: MAE, MAPE, RMSE, MSE, correlation coefficient, and R squared are 0.45%, 1.45%, 0.58%, 0.3%, 0.997%, and 0.998%, respectively. In essence, the HDL model can significantly augment the precision and trustworthiness of cost predictions throughout product development in comparison to existing techniques.