Research Article

Machine Learning and Reverse Methods for a Deeper Understanding of Public Roadway Improvement Action Impacts during Execution

Table 5

Number of disruptions by type for case study locations.

CaseEvent typeImbalanced trainingImbalanced testImbalanced TotalBalanced training

I-66 WCollision12,391 (0.45%)18,348 (0.85%)30,739 (0.62%)139,934 (14.01%)
Improvements188,979 (6.81%)77,751 (3.60%)266,730 (5.41%)188,979 (18.92%)
Both584 (0.02%)685 (0.03%)1,269 (0.03%)70,561 (7.06%)
None2,571,486 (92.72%)2,060,336 (95.51%)4,631,822 (93.94%)599,386 (60.01%)
Overall2,773,4402,157,1204,930,560998,861

I-81 SCollision1,846 (0.07%)2,279 (0.09%)4125 (0.08%)141,802 (16.78%)
Improvements36,859 (1.50%)33,707 (1.37%)70,566 (1.43%)143,867 (17.02%)
Both76 (0.003%)29 (0.001%)105 (0.002%)56,967 (6.74%)
None2,426,499 (98.43%)2,429,265 (98.54%)4,855,764 (98.48%)502,596 (59.46%)
Overall2,465,2802,465,2804,930,560845,232

I-64 WCollision25,171 (1.17%)3,854 (0.18%)29,025 (0.67%)138,654 (16.15%)
Improvements358,359 (16.61%)355,936 (16.50%)714,295 (16.56%)142,859 (16.64%)
Both9177 (0.43%)2,573 (0.12%)11,750 (0.27%)66,654 (7.76%)
None1,764,413 (81.79%)1,794,757 (83.20%)3,559,170 (82.50%)510,546 (59.45%)
Overall2,157,1202,157,1204,314,240858,713