Range-free indoor localization based on the Approximate Point-In-Triangle algorithm is widely used in lowcost wireless systems, but its reliability is strongly affected by radio irregularities and binary classification errors. In practical scenarios, the main limitation of classical APIT is not only accuracy, but also the strong variability of the results when the deployment area, neighbor distribution, and node mobility change. This paper proposes a probabilistic APIT formulation designed to improve both classification correctness and statistical stability. The method is based on Monte Carlo neighbor generation, RSSIgated selection, and soft inside/outside decisions with marginbased voting. Instead of producing a single deterministic output, the algorithm evaluates classification correctness and result variability under different geometric and radio conditions. Simulation and experimental results show that probabilistic APIT increases the number of correct classifications compared to classical APIT, while exhibiting a strong dependence on test area scale, anchor geometry, and neighbor spatial distribution. Real RSSI measurements collected with an ESP32-based indoor testbed confirm the same trends observed in simulation, showing that the proposed method improves classification reliability but remains sensitive to area scaling and deployment geometry.
