Performance Analysis of Distinctive Activation Functions in Stacked Denoise Autoencoder for Noise Removal in Vehicle Traffic Data

Vehicle Traffic Data is very crucial for accurate vehicle traffic flow prediction. Missing data and noise in raw data affect the quality of traffic-oriented processes such as vehicle traffic congestion, Route planning, Vehicle Distance Estimation, Road traffic Navigation etc. Deep Learning algorithms and their activation functions play a vital role in vehicle traffic-oriented applications. Stacked Denoise auto encoder has been used in many researches to remove noise and inconsistent data. In this paper, performance analysis of five activation functions is analyzed in a stacked denoise auto encoder to eradicate the noise and replace the missing values. Results are compared using performance metrics RMSE, and MAE. Exponential Linear Unit (ELU) activation executes well among five activation functions in traffic data and reduces the reconstruction error rate RMSE to 7.16 from 11.01 and MAE to 5.14 from 9.08 respectively.