Advances in micro-nano-biosensor fabrication are enabling technology that can integrate a large number of biological recognition elements within a single package. As a result, hundreds to millions of tests can be performed simultaneously and can facilitate rapid detection of multiple pathogens in a given sample. However, it is an open question as to how to exploit the high-dimensional nature of the multi-pathogen testing for improving the detection reliability a typical biosensor system. In this paper, we discuss two complementary high-dimensional encoding/decoding methods for improving the reliability of multi-pathogen detection. The first method uses a support vector machine (SVM) to learn the non-linear detection boundaries in the high-dimensional measurement space. The second method uses a forward error correcting (FEC) technique to synthetically introduce redundant patterns on the biosensor which can then be efficiently decoded. In this paper, experimental and simulation studies are based on a model conductimetric lateral flow immunoassay that uses antigen-antibody interaction in conjunction with a polyaniline transducer to detect presence or absence of pathogen in a given sample. Our results show that both SVM and FEC techniques can improve the detection performance by exploiting cross-reaction amongst multiple recognition sites on the biosensor. This is contrary to many existing methods used in pathogen detection technology where the main emphasis has been reducing the effects of cross-reaction and coupling instead of exploiting them as side information.