loading page

Momentum-Enhanced Linear Regression for Faster Convergence in Real-World Predictions
  • Hussein Al-Bazzaz
Hussein Al-Bazzaz

Corresponding Author:[email protected]

Author Profile


This paper investigates the efficacy of linear regression enhanced by gradient descent with momentum for predicting real-world outcomes. The introductory sections establish the significance of machine learning in driving sustainable innovations and detail the foundational aspects of linear regression, a statistical technique pivotal for modelling relationships between variables using a least squares approach. Enhanced with momentum-based gradient descent, these models achieve faster and more stable convergence, which is particularly beneficial in complex, real-life data scenarios. Through empirical analysis using the Boston and California housing datasets, we demonstrate that linear regression, when optimized, can effectively predict housing values with high R 2 scores, indicating robust predictive power across socioeconomic and geographic variables. Our findings underscore the model's utility as a forecasting tool in today's data-driven landscape. Future research directions include optimizing momentum coefficients and learning rates and potentially incorporating adaptive methods to enhance convergence efficiency. This study provides insights into the continuous improvements required in predictive analytics to maintain accuracy and reliability in diverse applications.
13 Apr 2024Submitted to TechRxiv
18 Apr 2024Published in TechRxiv