Reza´s algorithm

Estimation, interpolation, forecasting and modeling are common engineering methodologies used to fit models to specific historical data. Recently, artificial neural networks have proven to be good candidates for modeling incorporating historical data. Also, fuzzy methods simulate complex and unpredictable systems in real-world applications. Adaptive filters, time series and statistical methods are other conventional estimating methods. Most of these methods first assume a parametric model for the system and then try to optimize the parameters so that the output error is minimized. This methodology (parametric modeling) forces a fixed structure to the behavior of the system. This causes a limitation in functionality and performance of the estimation process as well as reducing the degree of freedom in general. However, Reza´s algorithm uses a sample of historical data, (X_i,Y_i, to estimate a value for Y₀ corresponding to a new value of X₀. It considers several possible ranges of solutions to calculate the final estimate of Y₀. Next, a candidate is introduced for each range of solutions (based on heuristic logic and engineering sense). Later on, these candidates are combined together using a weighted average method to reach a final solution for Y₀. This algorithm has been applied to several real-world data sets. The results show a relatively high accuracy (approx. 7%) in the sense of mean absolute percentage error. Reza is a sample-based algorithm, and in contrast to the other estimating methods, it does not force any parametric model to the system.