Web Request Predictions

Abstract

If one could predict future web requests, it would be possible to make the web much faster. One could fetch web resources before they are needed. When the human user clicks on a link, the needed data would already have been downloaded. We have created several algorithms that attempt to predict future web requests based on past histories. Our research evaluates and compares these prediction algorithms against real histories of web usage. Prediction algorithm results are compared based on correct predictions, erroneous predictions, and prediction rate. Some algorithms make predictions rarely but accurately, while others may predict more often but with less accuracy. To take full advantage of this, we combine multiple algorithms and use different voting strategies to determine the best prediction.