Short Time Series Forecasting: Recommended Methods and Techniques

This paper tackles the problem of forecasting real-life crime. However, recollected data only produced thirty-five short-sized crime time series for three urban areas. We present a comparative analysis four simple and machine-learning-based ensemble methods. Additionally, we propose five techniques that manage seasonal component series. Furthermore, used symmetric mean average percentage error Friedman test to compare performance methods proposed techniques. The results showed moving with removal produce best these It is important highlight high has no auto-correlation level symmetry, which deemed as white noise and, therefore, difficult forecast.