Learning Manifolds in Forensic Data

摘要: Chemical data related to illicit cocaine seizures is analyzed using linear and nonlinear dimensionality reduction methods. The goal find relevant features that could guide the analysis process in chemical drug profiling, a recent field crime mapping community. has been collected gas chromatography analysis. Several methods are tested: PCA, kernel isomap, spatio-temporal isomap locally embedding. ST-isomap used detect potential time-dependent manifold, being sequential. Results show presence of simple manifold very likely this cannot be detected by PCA. temporal regularities also observed with ST-isomap. Kernel PCA perform better than other methods, more robust when introducing random perturbations dataset.