Application of Machine Learning Algorithms for Visibility Classification

Visibility is one of the most important weather impacts on transportation systems. Due to the complexity of weather variables, classification and forecasting techniques for low visibility scenarios remain a challenge and a matter of interest and concern for transportation agencies nationwide. Florida is among the top-rated states with regards to safety issues due to low visibility and has suffered from multiple pileup accidents due to this issue. This paper presents an exploratory study of using machine learning (ML) algorithms to classify visibility conditions in three classes (i.e., low, moderate, and good) based on data obtained from weather stations, specifically in Florida. Five different ML classifiers were developed and spot-checked to determine the best suited for the visibility classification task. The classifiers showed promising results. The highest average accuracy score (89.71%) was achieved by an artificial neural network (ANN). The models were developed using Python.

[1]  Leslie F Musk,et al.  Climate as a factor in the planning and design of new roads and motorways , 1991 .

[2]  Hussein Almuallim,et al.  3 – Development and Applications of Decision Trees , 2002 .

[3]  P. Ray,et al.  Evaluation of Fog Predictions and Detection , 2015 .

[4]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[5]  Munindar P. Singh,et al.  Weather Forecasting Model using Artificial Neural Network , 2012 .

[6]  Sebastian Scherer,et al.  VoxNet: A 3D Convolutional Neural Network for real-time object recognition , 2015, 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

[7]  Robert L. Vislocky,et al.  An Automated, Observations-Based System for Short-Term Prediction of Ceiling and Visibility , 1997 .

[8]  M. Pagowski,et al.  Fog Research: A Review of Past Achievements and Future Perspectives , 2007 .

[9]  Lukasz Romaszko,et al.  Signal Correlation Prediction Using Convolutional Neural Networks , 2015, Neural Connectomics.

[10]  Gérard Dreyfus,et al.  Single-layer learning revisited: a stepwise procedure for building and training a neural network , 1989, NATO Neurocomputing.

[11]  Ferenc Wantuch.  Visibility and fog forecasting based on decision tree method , 2001 .

[12]  P. S. Patheja,et al.  An Enhanced Approach for Weather Forecasting Using Neural Network , 2011, SocProS.

[13]  Caren Marzban,et al.  Ceiling and Visibility Forecasts via Neural Networks , 2007 .

[14]  Antonello Pasini,et al.  A neural network model for visibility nowcasting from surface observations: Results and sensitivity to physical input variables , 2001 .

[15]  Mohamed Abdel-Aty,et al.  A Comprehensive Investigation of Visibility Problems on Highways: Developing Real Time Monitoring and Prediction System for Reduced Visibility and Understanding Traffic and Human Factors Implications , 2015 .

[16]  S. Sujitjorn,et al.  An expert system to forecast visibility in Don-Muang Air Force Base , 1994, Proceedings of IEEE International Conference on Systems, Man and Cybernetics.

[17]  Joby L. Hilliker,et al.  An Observations-Based Statistical System for Warm-Season Hourly Probabilistic Forecasts of Low Ceiling at the San Francisco International Airport , 1999 .

[18]  Silas Michaelides,et al.  Probabilistic Visibility Forecasting Using Neural Networks , 2007 .

[19]  Melissa K. Peterson.  Updates to the Changes Included in Amendment 78 to Annex 3: Meteorological Service for International Air Navigation , 2018 .