Search engine trend forecasting

Abstract

Time series analysis, forecasting and control are incredibly vast domains which have been put to use throughout a large number of areas, both industrial and educational. These areas include fields such as economics, business, engineering, the natural sciences and social sciences. Throughout this study we look at a modelling technique known as the Box-Jenkins methodology used together with the ARIMA models. We also look al dimensionality reduction and time series comparison through the use of SAX. These two techniques are incorporated and used together with the Google Ngram dataset (part of the Google Books project) in order to provide insight into the similarities between a time series and the forecasts that it produces. We start by explaining all the background information and know ledge required for the complete understanding of this study. We then describe both techniques extensively, reviewing the literature in their respective fields, explaining pros and cons of various approaches. We then move on to define the specifications, design and implementation techniques used in order to produce a working artefact (including the structure and use of the Google Ngram dataset), once again giving an in-depth depiction of how an adaptation of the Box-Jenkins methodology and ARIMA were used together with SA..X. We will then show how the artefact uses both methodologies and techniques together in order to facilitate the comparison of different time series, their models and their forecasts. Finally we show the correctness of our approach through the use of a sample data set taken from the original Google Ngram dataset. This is used to display various forecast functions and models, together with their comparison through the use of the SAX discretization methodology.