The spread of IOT based wearable equipment has made Personalized fitness recommender systems increasingly used and have become important for providing personalized assistance by collecting user-driven data. In the given research work, we have introduced a better multi-objective based personalized fitness recommendation system utilizing a novel three-tier deep learning approach. The proposed method implements a multi-objective method, where various goals like prediction of accurate distance, optimizing speed sequence, and personalized heart rate modeling are parallelly taken care of to deliver all-inclusive and effective fitness recommendations. The proposed three-tier Deep learning architecture includes the Convolutional Capsule Networks (ConvCaps), Bi-LSTM and Recurrent Neural Network (RNN) layers, designed to sequentially model related inputs and time-series trial data, qualifying innate feature extraction and effective performance. Experimental evaluations on real-world Fitbit fitness datasets exhibit significant improvements in predicting the accuracy, effective personalized recommendation, and results are as per user personalized manner compared to the existing models. This methodology emphasizes the effectiveness of multi-objective deep learning approaches in progressing the precision and personalization of Fitbit fitness recommender systems.