Development of machine learning models for categorisation of Nigerian Government's procurement spending to UNSPSC procurement taxonomy

Public procurement spending in Nigeria are usually documented and presented in non-standardised formats. This manifests in spends categorisation and classification inefficiencies. To address this, this research uses natural language processing (NLP) to classify the government's procurement spending based on the United Nations Standard for Product and Service Code (UNSPSC) procurement taxonomy. This research developed a machine learning model for the classification of procurement spending to the UNSPSC commodity level. The dataset was obtained from federal procuring entities. TF-IDF was used to transform them into NLP features. Multiple machine learning algorithms were employed to develop the classification model. The best performing algorithm is SVM with a 93% and 92% accuracy under the train-test split and k-fold cross-validation respectively. The higher level of accuracies obtained for many of the algorithms mean that the model can be practically deployed for the classification of the procurement spending based on UNSPSC standard procurement taxonomy.