This paper extends the analysis of a statistical methodology for source detection and localization (SDL) in a network of randomly distributed wireless nodes equipped with homogeneous and omni-directional sensors. The investigations are focused on SDL with respect to the nearest sensor node and are based on the observed source (phenomenon) energy. In this framework, the SDL algorithms are viewed as classification problems which are solved using pattern recognition techniques. In the presented approach: (i) sensors are randomly distributed and little is known about their exact locations; and (ii) a self-calibrating mechanism is proposed for creating the dataset whose feature vectors constitute the reference points for sensor locations in the space of sensor readings. The performance of the proposed algorithms is evaluated through Monte Carlo simulations and is demonstrated to be robust in the presence of noise and changes in the propagation environments.