Análise dos dados obtidos pelos sensores de trânsito através de séries temporais

Abstract

The evolution of digital technologies allows techniques and tools for monitoring vehicle traffic to become more efficient. This is why, it would not be viable for a human operator to deal with so much information alone. However, problems can occur while capturing information and damaging it. This erroneous or insufficient information, which does not follow a standard behavior when compared to the behavior of most data, is called anomalies. And, for the detection of these unexpected objects, time series can be used. A series of observations made sequentially over time are called time series. Knowing this, the objective of the work is to analyze the data captured by the traffic sensors through time series. In its development, fundamental theoretical concepts for understanding the methodology, identification of anomalous patterns in the database and application of prediction methods such as: Arima, Prophet, Simple Exponential Smoothing, Walk Forward and Random Walk. Finally, the methods were compared using error metrics: RMSE and MAE. In summary, it was observed that the Prophet prediction method performed better than the other techniques.