During aging, concretes and in overall geomaterials are weaken because of, for example, water penetration or increase of microcracks. Conventional methods to evaluate such weaknesses are coda wave interferometry with pairs of stations and ultrasonic pulse velocity test. This study aims to evaluate feasibility of machine learning methods for using single station coda wave interferometry. To this end, a numerical model of a 2 m concrete beam with highly scattering medium was considered with crack density of 0.1. To induce weakness a reduction in velocity was considered in a randomly determined depth of the concrete beam. Results show that if a continues monitoring is available, machine learning methods could predict the location and the amount of the weakness with an accuracy more than 99 and 95%.