1 A comparing performance of linear regression and radial basis function models for predicting student achievement Haviluddin, A. Sunarto, Suci Yuniarti haviluddin@gmail.com, andang99@gmail.com, mbakyu_niarti@yahoo.co.id Abstract This paper presents an approach for student achievements characterization data using statistical and artificial neural networks (ANN), namely linear regression and radial basis function neural network (RBFNN) methods at Islamic University, Bengkulu, Indonesia. The results of measurement were then compared to the value by the mean of square error (MSE). The experiment results showed that MSE 0.076 with model Y = 3.193 + 0.002 for linear regression and MSE 0.003, model Y = (1)T + (0.0021) with sum-squared error goal 0.01, and spread 1 for the RBFNN. Therefore, the smallest MSE value indicates a good method for accuracy, while RBFNN finding illustrates proposed best model to analyze student achievement characterization data. Keywords: linear regression, ANN, RBFNN, MSE, student achievement 1. Introduction Researchers widely and mostly employ the analysis model that combines statistical methods and artificial neural networks (ANNs). Thus, the use of these combined methods broadly used in finance, demography, and weather. In analysis scheme, the use of combination methods is influenced by the pattern of analysis models of particular data because each method has different sequences. The analysis model results that have high accuracy and good are very significant in decision-making; for example for predicting, designing and creating (Donate, Sanchez, & Miguel, 2012; Valipour, Banihabib, & Behbahani, 2013). Hence, the main factor affecting the selection of analyzing model techniques is to identify and determine based on the data characteristics. Time-series is part of data characteristics that consist of four features. They are (1) trend (T) that data have characteristics that tend to go up and down; (2) seasonal variation (S) means that type of data in the annual periodic fluctuations such as month, week and day; (3) cycles (C) refers to periodic fluctuations in the data type more than one year and; (4) random component (R) means a type of time series which is a combination of seasonal variation, trends, cycles and random factors (Fu, 2011; Santos, Silva, Silva, & Sene, 2011; Wei, 2006). Furthermore, a time series consists of a set of observations based on time. In principle, time series are used to analyze a set of data (y t +1 , y t +2 , ..., y t-n ) based on the data (x t +1 , x t +2 , ..., x t-n ) in a certain time frame (Box, Jenkins, & Reinsel, 2008; Donate et al., 2012; Zhang, 2003). In this paper, linear regression statistical and RBFNN have been used to analyze model using trend data characteristics. This article consists of four sections. Section-1 is the motivation to do the writing of the article. Next, the literature related to the theory and techniques in forecasting time series is discussed in Section-2. Section-3 presents the experimental results, and finally Section-4 describes the discussion results and research summaries. 2. Literature Review Analyze model is the first activity of modeling what will happen in the next stage based on previous data. In this paper, we conducted an experiment using the linear regression and RBFNN methods to analyze a model. Next, we describe briefly about linear regression and RBFNN technique. 2.1. Linear Regression The linear regression is a statistical analysis technique that is used to describe the causal relationship between the response (Y) variable and one or more explanatory (X 1 ,X 2 ,..,X N ) variables. Furthermore, the