Detection of anomalous transactions in a sugar company using Machine Learning models

The early detection of anomalous transactions is critical for preventing fraud and enhancing financial control within the agro-industrial sector. This research presents the design and implementation of a Machine Learning-based system for anomaly detection in the transactions of a Peruvian sugar company over the period 2022-2024. Unsupervised learning algorithms such as Isolation Forest, Local Outlier Factor, One-Class SVM, and Autoencoder were applied, and performance metrics for anomaly detection were compared, with One-Class SVM being the highest-scoring algorithm. The results highlight the potential of artificial intelligence to strengthen continuous audit processes and improve operational reliability in the Latin American agroindustrial context.