|
Hi, I have got a relatively small data set where each vector contains up to 800 features. The first few features contain date-related features (year, month, day, day of the week, nth occurrence of the day in the month, that kinda stuff). Each vector is data from a single day. The rest are simply numbers. Some of these numbers largely depend on the date-related features, but the majority are influenced on the preceding vectors. Which is influenced by what is largely unknown (of course this could probably be calculated on a per-feature basis using regression, but that is not worth it really). Now is my question: What are my options in order to try and predict complete future vectors (where the future can be 1 day ahead to, say, 2 weeks ahead)? Any suggestions for approaches / algorithms? The input for available for such algorithms would be the date-related features of the day for which the vector is to be predicted and the vectors of N preceding days. My background is in machine learning and artificial intelligence in general. However, that mostly concerned classification problems, which seems less related to this. |
