Anomaly Detection in GeoLLM Using Variational Autoencoder

Geospatial-based Large Language Models are an emerging technology based on the geospatial locations and the different applications based on the locations. From the survey, it is seen that the GeollM produces the misinformation for the requested input due to various factors. The accuracy of the response needs to be high and the LLM should not provide the misinformation regarding the location of the required output. Thus, this paper proposes the anomaly detection of factors that pay off the misinformation using the Variational Autoencoder in the GeoLLM system. The Variational Autoencoder provides the context-based information from the attention model, where the Linear Regression model is used to find the anomalies that are the reason for the misinformation in the GeollM system. The simulation carried out for the proposed work, the anomalies based on the parameters chosen are calculated, and the metrics for the proposed method.