Word analysis prediction based on ARIMA model

Abstract

First, Data processing was first performed to judge outliers if they were outliers, and finally missing values were filled in, and for the remaining valid data, change trend graphs were drawn to judge the trend of change in ascending rates and to predict future trends, using three mathematical models, ARIMA model, principal component analysis and Lasso regression model, which were used to describe the past change trends, and predict the level of change in the number of future word predictions. A comparative analysis of the models used above was conducted to compare the accuracy of each model, to derive which model predicted the most accurately, to give a rationale for the analysis, and to compare the accuracy of each model. Second, a model was built to analyze whether the change in the number of correct word predictions was related to which attributes of the words themselves and to elaborate the relationship. The data were subjected to KMO and Bartlett's tests to determine if a principal component analysis could be performed. This is followed by an exploration of whether the attribute has an effect on the number of correct predictions, how the change is affected, a visual presentation in the form of a graph, and a conclusion. To present an opinion, the paper considered the main factor that affects the number of predictions and propose measures to reduce the scenario brought by this factor. Finally, the paper came up a constructive report with listing and describing some other interesting features of this dataset.