Detecção de anomalias em um conjunto de dados de imóveis comerciais no município de Aracaju

Abstract

The present work uses the local anomaly factor (LOF) to detect anomalies in a data set of commercial properties, located in the city of Aracaju, made available by the Aracaju Real Estate Market Observatory (OMI-AJU), of the Municipal Finance Secretariat. from Aracaju (SEMFAZ). A well-known semi-automatic procedure for selecting the optimal values of the hyperparameters involved in the LOF calculation is applied. As the data are of mixed types, that is, quantitative and qualitative, the Gower distance is adopted to calculate the LOF. Six anomalies are identified. The anomaly with the highest LOF is a property with 14 floors, in the Centro neighborhood. There are three other properties, with 14 floors, in the same neighborhood. The values of the variables, referring to these three properties, are practically the same and are different from the one identified as an anomaly. This may explain the anomaly, suggesting that there is an error in the data relating to these three properties or that such data are repeated observations. The anomaly with the lowest LOF is a property, also in the Center, which is among the most expensive in the municipality, but has a much higher commercial density than that of this group of properties, which suggests an incompatibility between the value of the property identified as an anomaly and its density commercial. In addition to contributing to OMI-AJU/SEMFAZ, providing mechanisms for identifying anomalies that can negatively interfere with the quality and modeling of real estate data, this work presents a methodological contribution to the field of anomaly detection, which consists of using of the aforementioned semi-automatic selection procedure for LOF hyperparameters, combined with the Gower distance, to identify anomalies in mixed-type data sets.