Use of Machine-Learning and Load–Velocity Profiling to Estimate 1-Repetition Maximums for Two Variations of the Bench-Press Exercise

Abstract

The purpose of the current study was to compare the ability of five different methods to estimate eccentric–concentric and concentric-only bench-press 1RM from load–velocity profile data. Smith machine bench-press tests were performed in an eccentric–concentric (n = 192) and a concentric-only manner (n = 176) while mean concentric velocity was registered using a linear position transducer. Load–velocity profiles were derived from incremental submaximal load (40–80% 1RM) tests. Five different methods were used to calculate 1RM using the slope, intercept, and velocity at 1RM (minimum velocity threshold—MVT) from the load–velocity profiles: calculation with individual MVT, calculation with group average MVT, multilinear regression without MVT, regularized regression without MVT, and an artificial neural network without MVT. Mean average errors for all methods ranged from 2.7 to 3.3 kg. Calculations with individual or group MVT resulted in significant overprediction of eccentric–concentric 1RM (individual MVT: difference = 0.76 kg, p = 0.020, d = 0.17; group MVT: difference = 0.72 kg, p = 0.023, d = 0.17). The multilinear and regularized regression both resulted in the lowest errors and highest correlations. The results demonstrated that bench-press 1RM can be accurately estimated from load–velocity data derived from submaximal loads and without MVT. In addition, results showed that multilinear regression can be used to estimate bench-press 1RM. Collectively, the findings and resulting equations should be helpful for strength and conditioning coaches as they would help estimating 1RM without MVT data.

Keywords

• sports
• resistance training
• monitoring
• regularization
• neural network