Integration of Prophet Model and Convolution Neural Network on Wikipedia Trend Data

Abstract

Forecasting a time series is an ever growing area in which various machine learning techniques have been used to predict and analyze the future based on the data gathered from past. “Prophet” forecasting model is the most recent development in forecasting the time series, developed by Facebook. Prophet is much faster and simpler to implement than the previous forecasting model such as ARIMA model. Classification of forecasting output can be done by applying convolution neural network (CNN) on the outcomes of the Prophet model. To get higher accuracy with lesser loss, the method runs CNN with the best possible deep layers. The yearly, weekly, daily seasonality and trends could be realized by Prophet Model. The paper shows classification of these output based on the varying types of seasonality and trends. The labeled output can then, train and test all the trends’ result and find out the accuracy and loss incurred in a CNN model. Applying different depth and parameters of CNN that is a combined unit at each layer, it can achieve more than 96% accuracy with less than 4% loss. The integration of prophet and CNN shows that the training and testing model of a neural network can validate the prediction done by using prophet forecasting model along with the seasonality and trends parameters are in coherence to one another.

Gymnastic Movement Recognition Based on Depth Convolution Neural Network

“…The weight sharing of the network model of the CNN means that all neurons on the same feature map are trained with the same weight parameters, as shown in Figure 2 [13][14]. If the image contains 100 feature maps, the total number of convolution kernel parameters that need to be trained is 100×100=104, which is 108 times lower than the 1012 parameters of the fully connected network, and the local connection and weight sharing are extremely large It reduces the computational complexity of parameters, making it possible to train and recognize images in deep networks [15][16].…”

Predicting Customer Behavior Using Prophet Algorithm In A Real Time Series Dataset

“…Two most used trend model implementations are saturating growth model and piecewise linear model (Sahay & Amudha, 2020). The saturating growth model is described by the basic equation (2):…”