Classification of urban soils for forensic purposes using supervised self‐organizing maps

Abstract

Due to its transferability, soil has been commonly used as evidence in criminal investigations. In this work, urban soil samples taken from five parks are subject of analysis. Infrared spectroscopy was used as experimental technique. Using this technique, a total number of 80 samples of urban soils were examined. The infrared spectra were properly pre‐processed and were further used for development of classification models based on supervised self‐organizing maps. In order to perform the variable selection and the optimization of the supervised self‐organizing maps in automated manner, genetic algorithms were employed. Using this approach, for three locations with our models, we are able to successfully predict all samples used for external validation, while for the remaining two locations, 96% of the samples, used for external validation, were successfully classified by the developed models. In addition to this, the developed models were used for examination of influence of the weather and the seasonal changes on the composition of the soil. For this purpose, 3 years after the initial soil samples were collected, additional 12 samples were analyzed. All these samples were correctly classified by the previously developed model.

X-ray Powder Diffraction and Supervised Self-Ogranizing Maps as Tools for Forensic Classification of Soils

“…[52][53][54] In our work, we use GAs not only as a variable selection tool but also in order to find optimal parameters of the developed models. 40,55,56 Genetic algorithms were also implemented in Matlab programming language. For this purpose, Genetic Algorithm Toolbox developed at University of Sheffield was used.…”

“…In our previous work, when we developed different models for classification of the urban soils based on infrared spectroscopy, the samples from these two parks were most often misclassified, probably due to smaller difference in the composition of the soils on these two lo-cations. 40 In this case, the sample D41 is misclassified as a sample from Moša Pijade High School Park (label: E). However, the second of these samples (label: E11) is classified together with the samples taken from the park which is in the neighborhood of the Colorful Mosque (labels: C).…”

“…Compared to the results which we obtained using infrared spectra of the urban soils, where only one sample was misclassified, we have to state that, in that case, due to higher overlap between signals in the infrared spectra there we were not able to develop good classification models which will be able to classify all five urban soils samples. 40 In that case, as stated earlier, we were forced to use one-against-the-rest approach and, consequently, we developed five separate models in order to obtain good classification models for all five locations.…”

“…[30][31][32][33][34][35][36][37][38][39] In our previous work, we successfully developed models for classification of urban soils for forensic analysis from five locations using infrared spectroscopy as an experimental technique. 40 However, the signals (the bands) in infrared spectra are highly overlapped most of the time. That was the reason why, in order to obtain successful classification of the samples, we used one-against-the-rest approach.…”

“…The performances of the one-against-the-rest approach were considerably better compared to a single model approach used for classification of all five types of urban soil samples. 40 Compared to the signals in the infrared spectra, the signals obtained by X-ray powder diffraction are less overlapped. Having this information in mind, in this work we decided to use X-ray powder diffraction as an experimental technique for classification of samples from nine locations.…”