Kidney failure is a critical chronic disease, defined as the irreversible loss of kidney functions. It has been shown that this pathology is associated with an increase of ammonia concentration in breath. Measuring it with a handheld system is a simple way for a noninvasive and early diagnostic. The idea of this paper is to measure the concentration of ammonia in a concentration range of human breath (500 ppb-2100 ppb) with humidity using a network of 11 different nanocomposite sensors. To overcome sensor weaknesses (sensor drift and sensitivity to humidity), the electronic nose principles are applied. Polyaniline-based nanocomposites with titanium dioxide, chitosan and carbon nanotubes are used to provide different sensitivities and response times and thus associate a single pattern to a concentration range. Several classifiers are then investigated and recursive feature elimination algorithm are used to select the most relevant features and sensors while improving the measurement accuracy. Diagnosis accuracy reaches 91% with the combination of feature selection and Support Vector Machine algorithm.