Filtering harbor craft vessels' fuel data using statistical, decomposition, and predictive methodologies

Filtering is the process of defining, recognizing, and correcting flaws in given data so that the influence of inaccuracies in input data on subsequent studies is minimized. This paper aims to discuss the characteristics of some filtering methods from various topics. Wavelet transform and frequency (Fourier) transform are considered for the decomposition methodologies whereas descriptive statistics is used for the statistical methodology. The Kalman filter and autoencoder neural network are also explored for the predictive methodologies. All the aforementioned methodologies are discussed empirically using two metrics of R-squared and mean absolute error. This paper aims to study the effectiveness of these filtering techniques in filtering noisy data collected from mass flowmeter reading in an unconventional situation i.e., on a tugboat while in operation to measure fuel consumption. Finally, the performance of various filtering methods is assessed, and their effectiveness in filtering noisy data is compared and discussed. It is found that the Haar wavelet transforms, Kalman filter and the descriptive statistics have a better performance as compared to their counterparts in filtering out spikes found in the mass flow data.