Predicting Habitual Activities for Reducing Perceived Latency

Human activities are usually habitual, following some patterns. The patterns make these activities predictable. A number of our activities these days are online, and these activities require us to access resources from online sources, incurring data transfer latencies. Being able to predict the patterns in online activities allows automatic prefetching of these resources so that the perceived latency of accesses can be vastly reduced. While there is a lot of prior research to predict short-term activities, research relating to predicting long-term habitual activities is sparse. This paper attempts to address the general area of how long-term habitual activities can be predicted. It proposes a model using long short-term memory (LSTM) capable of learning long-term habitual activities and predicting future activities based on this model. Our experimental evaluation shows that the prediction accuracy is promising.