Current inference techniques for processing multi-needle Langmuir Probe (m-NLP) data are often based on the Orbital Motion-Limited (OML) theory which relies on several simplifying assumptions. Some of these assumptions, however, are typically not well satisfied in actual experimental conditions, thus leading to uncontrolled uncertainties in inferred plasma parameters. In order to remedy this difficulty, three-dimensional kinetic particle in cell simulations are used to construct synthetic data sets, which are then used to train and validate regression-based models capable of inferring electron density and satellite potentials from 4-tuples of currents collected with fixed-bias needle probes similar to those on the NorSat-1 satellite. Based on our synthetic data, the techniques presented enable excellent inferences of the plasma density, and floating potentials, while the generally accepted OML inferred densities are approximately three times too high. The new inference techniques that we propose, are applied to NorSat-1 data, and compared with OML inferences. While both regression and OML based inferences of floating potentials agree well with synthetic data, only regression inferred potentials are consistent with satellite measured currents, indicating that the regression based inference models are more robust and accurate when applied to satellite data.