Time Series Prediction Using Pattern Matching

Time series forecasting is very important in several domains and has received a lot of interest from researchers in recent years. In this paper, we investigate the use of pattern matching technique in seasonal or trend time series prediction. First, this technique retrieves the sequence prior to the interval to be forecasted. Then this sequence is used as a sample for searching k-nearest neighbors or neighbors within a threshold T in historical data. Next, sequences next to these found patterns are retrieved. Finally, the forecasted sequence is calculated by averaging the sequences found in the immediate previous step. The experimental results showed that this approach produces competitive results on seasonal or trend time series in comparison to artificial neural network (ANN) in terms of prediction accuracy and time efficiency. In our experiment, we also examine the impact of parameter values k and T on the predictive accuracy.