The legacy mathematical equations for rate of penetration (ROP) calculation of hydrocarbon and geothermal wells are unreliable, primarily due to dependency of ROP on multiple interrelated and unrelated mechanical, geological, hydraulic and geomechanical variables. Therefore, we present a multiple linear regression machine learning (ML) method, developed and tested for reliable ROP prediction from measured depth (MD) and mechanical variables weight on bit (WOB), revolutions per minute (RPM), pump pressure (SPP) and torque (TQA). Exploratory data analytics is applied on depth series data of Utah Forge field (USA) geothermal well # 58–32 to obtain insights into the well construction operations. Simple and multiple linear regression is applied on 1 ft interval depth series data from 2173 to 7536 ft. The simple linear regression ML model of 8 ¾” phase with ROP as dependent and MD as independent variable gives R2 = 0.423, which indicates overall unpredictable variation of ROP with MD due to interlayered thin beds of hard and soft rocks, variations in lithology, several core jobs, unpredictable changes in drillability of rocks, drilling challenges and several bit changes. However, multiple linear regression model for ROP prediction of 8 ¾” phase reveals R2 = 0.757, which is reliable. Although there is no apparent relation of ROP with MD, WOB, RPM, SPP and TQA in this 5363 ft well section, multiple linear regression ML model provides good correlation of ROP. The reliability of this model can be further improved with inputs of lithology and drilling fluid data.