Accurate and online prediction of heat exchanger (HE) fouling is one of the primary requirements for precise control, predictive maintenance, and operational continuity. As fouling tends to alter the HE dynamics, a dual extended Kalman filter (DEKF) becomes the ideal technique to predict fouling along with the HE states concurrently. A modification in DEKF is proposed in this work to estimate the states of HE and fouling resistance (FR) using a linear parametric varying (LPV) model. FR prediction model of DEKF is restructured to include a machine learning (ML) model to provide guiding input. The guiding input provides a preliminary estimate of FR, which needs to be fine-tuned by the DEKF. This reduces the overhead on DEKF and enables faster convergence. GA is used to tune the weightage given to the guiding input from the ML model, which can improve the overall estimation accuracy. The performance of the proposed DEKF is comparatively evaluated under five different fouling conditions encountered by an industrial HE. Experimental results demonstrate about 38.49% improvement in estimation accuracy for FR on average.