Laser induced breakdown spectroscopy (LIBS) allows quick analysis to determine the elemental composition of the target material. Samples need little\no preparation, removing the risk of contamination or loss of analyte. It is minimally ablative so negligible amounts of the sample is destroyed, while allowing quantitative and qualitative results. Vacuum ultraviolet (VUV)-LIBS, due to the abundance of transitions at shorter wavelengths, offers improvements over LIBS in the visible region, such as achieving lower limits of detection for trace elements and extends LIBS to elements\samples not suitable to visible LIBS. These qualities also make VUV-LIBS attractive for pharmaceutical analysis. Due to success in the pharmaceutical sector molecules representing the active pharmaceutical ingredients (APIs) have become increasingly complex. These organic compounds reveal spectra densely populated with carbon and oxygen lines in the visible and infrared regions, making it increasingly difficult to identify an inorganic analyte. The VUV region poses a solution as there is much better spacing between spectral lines. VUV-LIBS experiments were carried out on pharmaceutical samples. This work is a proof of principle that VUV-LIBS in conjunction with machine learning can tell pharmaceuticals apart via classification. This work will attempt to test this principle in two ways. Firstly, by classifying pharmaceuticals that are very different from one another i.e., having different APIs. This first test will gauge the efficacy of separating into different classes analytes that are essentially carbohydrates with distinctly different APIs apart from one another using their VUV emission spectra. Secondly, by classifying two different brands of the same pharmaceutical, i.e., paracetamol. The second test will investigate of the ability of machine learning to abstract and identify the differences in the spectra of two pharmaceuticals with the same API and separate them. This second test presents the application of VUV-LIBS combined with machine learning as a solution for at-line analysis of similar analytes e.g., quality control. The machine learning techniques explored in this thesis were convolutional neural networks (CNNs), support vector machines, self-organizing maps and competitive learning. The motivation for the application of principal component analysis (PCA) and machine learning is for the classification of analytes, allowing us to distinguish pharmaceuticals from one another based on their spectra. PCA and the machine learning techniques are compared against one another in this thesis. Several innovations were made; this work is the first in LIBS to implement the use of a short-time Fourier transform (STFT) method to generate input images for a CNN for VUV-LIBS spectra. This is also believed to be the first work in LIBS to carry out the development and application of an ellipsoidal classifier based on PCA. The results of this work show that by lowering the pulse energy it is possible to gather more useful spectra over the surface of a sample. Although this yields spectra with poorer signal-to-noise, the samples can still be classified using the machine learning analytics. The results in this thesis indicate that, of all the machine learning techniques evaluated, CNNs have the best classification accuracy combined with the fastest run time. Prudent data augmentation can significantly reduce experimental workloads, without reducing classification rates.