Chapter VI
Machine Learning
79
VI.5.1. Database cleaning
Given that erroneous data can significantly compromise the performance of the models to be
built, the initial database was reduced by eliminating data classified as "unreliable" tests (RF=4)
or data where the cross-section was considered "very complex" (CF=4). Clear inconsistencies
within the database were detected and removed. All rows with missing values were deleted,
resulting in a reduced database containing 8091 tests.
Figure VI-7 Overview of the cleaned Database
VI.5.2. Application of Froude Similarity
Froude similarity is a method used to resize hydraulic parameters in order to achieve similar
behaviors between reduced-scale prototypes and real-world conditions. It is based on the ratio of
inertial and gravitational forces to maintain the same proportions and characteristics. The Froude
similarity law was employed to provide the best generalization of large-scale applications. The
equations (Equation VI-1, Equation VI-2, Equation VI-3 and Equation VI-4) give the necessary
laws for Froude similarity scaling.
𝑛 𝐿 = 1/H mo,toe observed
Equation VI-1
[𝑛 Hm0 , 𝑛 h , 𝑛 ht , 𝑛 Bt , 𝑛 B , 𝑛 hb , 𝑛 Rc , 𝑛 Ac , 𝑛 Gc ] = 𝑛 𝐿
Equation VI-2
𝑛 T = 𝑛 𝐿
0.5
Equation VI-3
𝑛 𝑞 = 𝑛 𝐿
1.5
Equation VI-4