top of page

Predicting Rheumatoid Arthritis Disease Activity with an Adaptive Deep Neural Network

Jul 23
1 min read

Rheumatoid arthritis (RA) is managed by tracking disease activity over time and adjusting treatment accordingly. This project builds a deep-learning model that predicts RA disease activity from clinical features, with the aim of supporting personalized, patient-specific treatment planning.

The approach

The model is an adaptive deep neural network built in TensorFlow/Keras — "adaptive" in that its architecture and regularization were tuned to fit the structure of the clinical data rather than fixed up front. Clinical features are cleaned and normalized, then fed through the network to estimate a patient's disease-activity level.

Technologies used

  • TensorFlow / Keras: define, train, and evaluate the deep neural network.

  • Deep neural network: learns non-linear relationships between clinical features and disease activity.

  • Adaptive architecture & regularization: network depth, width, and techniques such as dropout are tuned to the data to control overfitting.

  • scikit-learn, Pandas & NumPy: preprocessing, feature scaling, train/test splitting, and evaluation.

  • Hyperparameter tuning: systematic search over model settings to improve performance.

Impact

By estimating disease activity per patient, the model supports personalized treatment planning and more informed clinical decisions.

Tech stack: Python, TensorFlow, Keras, deep neural networks, scikit-learn, Pandas, NumPy, hyperparameter tuning.

 
 
 

Comments

bottom of page