An Optimized Machine Learning Model For Pavement Condition Prediction

• Author(s):

  Himanshu Kumar Pandey | Prof. Vinay W. Deulkar

• Keywords:

  Pavement Condition Prediction, Particle Swarm Optimization, Neural Networks, Back Propagation, Conjugate Gradient Back-propagation With Fetcher-Reeves Restarts, Classification Accuracy.

• Abstract:

  Pavement Condition Prediction Is An Essential Aspect Of Road Maintenance Planning, Ensuring The Safety, Efficiency, And Durability Of Transportation Infrastructure. Statistical Machine Learning (ML) Techniques Have Emerged As Powerful Tools For Addressing This Challenge, Offering Data-driven Methods To Model Complex Relationships Between Pavement Performance And Contributing Factors Such As Traffic Loads, Weather, And Material Properties. Traditional Prediction Models Include Linear Regression, Markov Chains, Mechanistic-empirical Approaches, And Time-series Models. Although These Methods Are Computationally Simple, They Often Exhibit Limited Accuracy Because Pavement Deterioration Is Highly Nonlinear And Influenced By Multiple Interdependent Factors. Recent Developments In Artificial Intelligence Have Enabled The Use Of Machine Learning Models For The Purpose. This Work Presents A PSO- Deep Neural Network Hybrid Model For Predicting Pavement Conditions. The PSO-Deep Neural Network Model Is Trained With Trained Using Conjugate Gradient Backpropagation (CGB) With The Fletcher–Reeves Restarts Offer Several Important Advantages Over Conventional Optimization Methods Such As Standard Gradient Descent, Such As Immunity To Vanishing Gradient And Local Minima Stagnation. The Model Attains An Accuracy Of 95.2% For Pavement Condition Prediction Outperforming Existing Work In The Domain.

---

Other Details

• Paper id:

  IJSARTV12I2104581

• Published in:

  Volume: 12 Issue: 2 February 2026

• Publication Date:

  2026-02-14

Download Article