<p>Critical Heat Flux (CHF) is a fundamental parameter for ensuring the performance, reliability, safety, and economic viability of water-cooled nuclear reactors. Despite its significance, the absence of a deterministic theory for CHF prediction poses a substantial challenge in thermal engineering. Consequently, various experimental and numerical models have been developed, yet no universally accepted model comprehensively addresses the wide range of flow conditions encountered in practical applications. The accurate prediction of CHF remains a complex and critical task. This work introduces a novel Deep Decoder Type Cascaded feedforward Neural Network (DD-CFNN) to predict CHF across a broad spectrum of operating conditions. The DD-CFNN architecture leverages the strengths of feedforward neural networks by incorporating connections from input and preceding layers to all subsequent layers, significantly enhancing its modeling capacity. Furthermore, the decoder type cascaded structure enables progressive enhancement of the feature space, optimizing its predictive performance. A new dataset, derived from extensive CHF data available in the literature, was developed for training and testing the proposed model. The performance of the DD-CFNN variants was evaluated extensively, with the best-performing model achieving a Root Mean Squared Error (RMSE) of 5.17% and a correlation coefficient of 0.997, indicating exceptional predictive accuracy. The model’s performance was benchmarked against standard machine learning algorithms such as Support Vector Machine (SVM) and k-Nearest Neighbors (KNN). Predictive robustness and reliability were further validated using various holdout and cross-validation techniques. The results demonstrate the superiority of the proposed DD-CFNN in predicting CHF and recommend its potential application as a robust monitoring tool to ensure the safe operation of nuclear power reactors.</p>