As the adoption of smart home technologies increases, the demand for accurate and cost-effective indoor positioning solutions also grows. Traditional high-precision systems often rely on beacons; however, beacon-less alternatives, such as visual positioning through high-definition cameras, face challenges including inconvenience and privacy concerns in residential environments. A promising solution to these challenges is the use of long-range capacitive sensors. These sensors, when coupled with regression neural networks to analyze variations in data from capacitive plates mounted on walls, have been proven to be both feasible and highly precise for determining a person's location within a space. However, the limited computational capacity of smart home embedded systems requires the models to be efficiently compressed without compromising their performance.