6EvaluationThis chapter is dedicated to the evaluation of the presented localization framework.Special attention is given to the description of the setting and the environment con-ditions of the experiments as this was often only briefly reported in the investigatedliterature1.After describing the experimental setup, the results for the training process of theradio propagation models are analysed. A focus is placed on the behaviour of thetraining process under different levels of 3D geometry granularity. An additionalexperiment for the radio propagation context is conducted by evaluating the abilityof the models to generalize over multiple devices. The generated models and the cor-responding observations are subsequently used for the evaluation of the localizationalgorithms that were designed and implemented.It will be reported how the three algorithms, the LMSE, the HMM and the PF, per-form on a selected fully trained propagation model. In the first step, the algorithmsare compared on synthetic data obtained from the propagation model under differ-ent noise levels. The differences between the results of the individual algorithmsare explained by their underlying algorithmic properties. The use of synthetic datamakes this task easier.Using the results of the radio propagation evaluation and the experiences from theanalysis on synthetic data enables a thorough investigation of the framework perfor-mance on real world conditions. Location annotated RSSI readings for a distance ofabout 8000 m were manually collected for eight different track configurations of vari-ous complexity. This evaluation corpus is used to estimate the localization errors ofthe three algorithms over differently trained radio propagation models. Furthermore,the effect of the devised Device Adaptation scheme described in 4.2.3, is analysedand interpreted.1Although it should be emphasized, that the investigated literature was mostly available in theform of journal papers which have surely made compromises due to page count constraints.