The use of Multivariate Statistics for Classification of Complex Mixtures in Forensic Aanalysis

Noa Cohen, DIFS, Israel Police , Jerusalem, Israel
Yaniv Avissar, DIFS, Israel Police, Jerusalem, Israel
Dan Muller, DIFS, Israel Police, Jerusalem, Israel
Arnon Grafit, DIFS, Israel Police, Jerusalem, Israel

As of today, single donor DNA analysis is the only form of evidence forensic scientist and others in the criminal field can trust as accurate. All other forms of evidence have some degree of uncertainty. Collaboration between the Forensic and the Technology divisions of Israeli police aims to develop statistical programs forensic scientist that can use to quantify the level of confidence forensic expert reports.

GCMS application enables separation and identification of compounds present in the sample. The common technique involves pattern recognition profile of the chromatogram by visual comparison against the known target compounds. Forensic chemists have to face difficulties during the isolation of residues of accelerants in the samples.

In order to compensate the challenges involve with analysis of chemometric information obtained by Gas Chromatography-Mass Spectrometry (GC-MS), a statistical model has been developed to support and validate the subjective interpretation of the analyst.

Multivariate statistical methods offer powerful tools in analyzing such complex data. The experimental data was received from the arson investigation lab of the police of Israel. 300 samples of gasoline and diesel collected from fire scenes were analyzed by GC-MS. The percent peak area of each of the compounds of total ion chromatogram obtained for each sample was extracted and represented the input variables.

Peak alignment algorithm was applied as an initial preprocessing step prior to the analysis, in order to correct shifts caused by retention time variation among the samples. The algorithm ensures that each point along the time axis in each chromatogram is registered in the same variable in the data matrix.

The data was divided to training set and test set. Different multivariate statistics have been applied including supervised learning and a-parametric methods. The results show correct classification of the samples according to their type.

We believe that the use of statistical technique could be a helpful tool for analyzing differences in complicated mixtures in other domains in the forensic analysis and can be extended from forensic science to a large range of applications including industrial food and flavors, perfumes and other natural extracts.

Forensic evidence should be one of the most objective and valid forms of testimony in criminal or civil law case. However, in the past few decades, wrongful convictions have been made and criminals have walked free because of a struggle to collect, analyze and interpret evidence. The international community demands stronger scientific foundations in forensics for the pursuit of truth and justice – and the Israeli police is responding with solutions.