Tieng Wei Koh

Abstract Large Language Models (LLMs) are increasingly being applied to time-series forecasting, giving rise to a class of models referred to as time-series LLMs. While these models achieve competitive predictive accuracy, their reliability and structural consistency remain insufficiently understood. These models may produce forecasts that are numerically accurate on average yet statistically or temporally implausible, a phenomenon referred to as hallucination.

5.2 Experiment of forecasting the number of crimes hourly The results of forecasting two models (ES-LSTM and ES-BiLSTM) using two different methods will be compared in this section. The first method uses the data for the number of crimes without including the number of crime types with the model (ES-BiLSTM). The second method uses the two models with the inclusion of the number of crime types to predict the total number of crimes with the models (ES-LSTM and ES-BiLSTM).