Scholarlink

<p>As fitness recommender systems evolve, individual interest-based recommendations have gained immense value, supported by enhanced data analytics collected through wearable devices. This work focuses on developing a novel system that leverages self-attention-driven Gated Recurrent Unit to interpret individual predictions of speed in the sporting activity along- with heart rate sequences to achieve their health goals. Existing research like FitRec targets either real-time estimation or one-way temporal modelling, whereas the proposed method incorporates attention mechanism to dynamically balance temporal features and makes use of tensor decomposition to learn the latent user embeddings. In the proposed methodology, the attention-MLP forecasts personalized workout distances correlated with calorie goals, then GRU predicts sport-specific speed and heart rate sequences. This two-prediction model allows accurate and responsive recommendations based on user’s objective such as weight reduction, endurance improvement and cardiovascular fitness. The system devised in this work offers exercise- distance prescriptions specific to individual calorie burn goals of each of the participants. It provides specific coaching to facilitate a continuum of fitness objectives, be it to reduce weight, increase stamina or general wellbeing sustenance. Individual training proposals are made with the constant modification of the key variables (selected routes, individually available training preferences, and contingent variables that influence the design of a personalized proposal. Besides, it can be said that the scope of the system is bigger than generic exercise advice because it produces personal heart-rate patterns that help them to better comprehend their heart-rate trends and to make health-based alterations which will lead to health-wise outcomes in the long term. To determine the performance of the model, real time user data were used in evaluating how well the model operates in predicting its dynamic outcomes as compared to the actual activity behaviors occurring. Further experiment, such as an ablation analysis, measured the performance of each module whereby the framework achieved absolute error levels of 2.02 on sport speed and 10.55 on heartrate prediction as part of the proposed fitness recommendation system.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Attention-driven time series prediction for personalized fitness recommendation using gated recurrent unit: a fusion technique

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Naguru Naveen Kumar Reddy,
  • Pedumoori Bhaja Govindu,
  • Bhuvaneswari Swaminathan,
  • Subramaniyaswamy Vairavasundaram

摘要

As fitness recommender systems evolve, individual interest-based recommendations have gained immense value, supported by enhanced data analytics collected through wearable devices. This work focuses on developing a novel system that leverages self-attention-driven Gated Recurrent Unit to interpret individual predictions of speed in the sporting activity along- with heart rate sequences to achieve their health goals. Existing research like FitRec targets either real-time estimation or one-way temporal modelling, whereas the proposed method incorporates attention mechanism to dynamically balance temporal features and makes use of tensor decomposition to learn the latent user embeddings. In the proposed methodology, the attention-MLP forecasts personalized workout distances correlated with calorie goals, then GRU predicts sport-specific speed and heart rate sequences. This two-prediction model allows accurate and responsive recommendations based on user’s objective such as weight reduction, endurance improvement and cardiovascular fitness. The system devised in this work offers exercise- distance prescriptions specific to individual calorie burn goals of each of the participants. It provides specific coaching to facilitate a continuum of fitness objectives, be it to reduce weight, increase stamina or general wellbeing sustenance. Individual training proposals are made with the constant modification of the key variables (selected routes, individually available training preferences, and contingent variables that influence the design of a personalized proposal. Besides, it can be said that the scope of the system is bigger than generic exercise advice because it produces personal heart-rate patterns that help them to better comprehend their heart-rate trends and to make health-based alterations which will lead to health-wise outcomes in the long term. To determine the performance of the model, real time user data were used in evaluating how well the model operates in predicting its dynamic outcomes as compared to the actual activity behaviors occurring. Further experiment, such as an ablation analysis, measured the performance of each module whereby the framework achieved absolute error levels of 2.02 on sport speed and 10.55 on heartrate prediction as part of the proposed fitness recommendation system.