A Neural Network Model of Students English Abilities

Paper:

A Neural Network Model of Students English Abilities Based on Their Affective Factors in Learning
Fitra A. Bachtiar , Katsuari Kamei , and Eric W. Cooper
School of Science and Engineering, Ritsumeikan University 1-1-1 Noji Higashi, Kusatsu, Shiga 525-8577, Japan E-mail: tra@spice.ci.ritsumei.ac.jp College of Information Science and Engineering, Ritsumeikan University 1-1-1 Noji Higashi, Kusatsu, Shiga 525-8577, Japan E-mail: {cooper@is, kamei@ci}.ritsumei.ac.jp [Received September 15, 2011; accepted November 15, 2011]
Graduate

The gap between teaching perspectives and students differences may impact negatively on teaching and learning effectiveness, indicating the need for a new approach for bridging this gap. The potentials of articial neural networks for approximating extremely complex problems encouraged us to develop an estimation model of student English ability. The model was trained using a back propagation algorithm and tested using 154 samples from two universities. The model estimation rate related to student English ability demonstrated a high level of estimation by 93.34% for listening, 94.38% for reading, 94.90% for speaking, and 93.58% for writing.

Keywords: neural network, affective factors, English ability, estimation model

1. Introduction
Teaching is considered to play a vital role as the formal medium for learning competence on the part of students [1]. Considering that effective teaching can facilitate learning, teaching is critical in the formation of knowledge, skills, and attitude. Low-quality teaching could result in unsuccessful learning. In fact, according to data in the Teaching Perspective Inventory by Pratt and Collins [2], over 90% of teachers hold only one or two perspectives as their dominant view in teaching and teachers perspectives vary with their views of knowledge, learning, and teaching. Most instructors, moreover, use a one-size-ts-all approach when teaching [3]. This approach could be misguided when applied to teaching practices. The one-size-ts-all approach is also not appropriate because individual students are naturally different from others in their cognitive and affective characteristics. Related to the affective factor, students may have different levels of motivation, different attitudes toward teaching and learning, and different responses to a specic classroom environments and instructional practices [3]. In terms of learning English, for example, some students are Vol.16 No.3, 2012

highly motivated because they want to work in an international company. The concept of affective factors described by Ellis [4] and Brown [5] clearly outlines the importance of affective factors in the learning of a language. These factors include, among others, motivation, attitude, inhibition, and anxiety. Notably, Williams and Burden [6] also emphasize the importance of student psychological factors for teachers to realize in language learning. More importantly, on a highly conceptual level, Immordino-Yang and Damasio [7] also argue the relevance of affective and social factors in learning. They put affective factors, rather than cognitive factors, as having a primary role in student learning. They argue that aspects of cognition are affected by and placed under the processes of emotion, which they term emotional thought. Affective factors play a role as a rudder that guides student rationality in mobilizing their cognitive potential to result in more rational actions. Attempts thus far to systematically maximize the benets of affective factors in education have not unfortunately been entirely successful. Filling in the gap between teaching perspectives and student diversity by examining and estimating student ability in affective factors could be a guide to teaching effectiveness. Estimating student ability can be utilized to reveal the strengths and weaknesses of a particular student. As has been suggested by Felder and Brent [3] in areas of learning, understanding student differences is a process of characterizing students in their affective factors. Identifying student characteristics in advance is a benet for the instructor in a number of important ways, such as selecting teaching materials, applying appropriate teaching methods and strategies, and determining learning resources and teaching media best suited for students. This paper presents a neural network model for estimating the English ability of students based on affective factors in learning. Related work is presented rst, followed by a brief overview of affective factors, and the three major factors of motivation, attitude, and personality are dened for application in our study. Data collected on student affective factors and the reduction of dimensions of data by factor analysis are described and the estima375

Journal of Advanced Computational Intelligence and Intelligent Informatics

Bachtiar, F. A., Kamei, K., and Cooper, E. W.

tion model is proposed. Finally, our experiment results and conclusion, including directions for future work, are presented.

2. Related Work
Research dealing with the social sciences in studying student data has been performed by a number of researchers. In the eld of affective factors in language learning, an attempt to reveal which type of motivation will have an inuence on specic motivations and other factors that can increase motivation has been carried out by Obeidat [8]. Wei [9] also studied the interrelatedness between motivation and anxiety. Halpern [10] investigated linguistic, cognitive, and affective factors that impact on English language learners performance in a reading tests. The results of research by these researchers give an insight into the important aspects underlying students internal factors and the interrelatedness among individual factors in language learning. Advanced research in predicting student performance using algorithms also has been carried out. Kumar et al. [11] applied data mining using k-means clustering to group data having the same features, and using decision trees for pattern analysis. A prediction method has also been reported using Neural Networks (NN) to classify students graduation outcomes at a 2-year institution, inferring which students would graduate successfully [12]. In this study, 12 parameters in student proles were used to predict graduation outcomes. A series of tests and experiments were performed to get the best average prediction rate of 77% on test data. Based on a study by Ibrahim and Rusli [13], NNs also outperform decision trees and linear regression in predicting students academic performance. NNs have been widely used in areas of prediction. The wide range of applications of NNs in many elds and sectors is due to their power to model behavior to produce an approximation of given output [14]. Our study has also been motivated by the potential use of NNs in yielding estimations for producing output based on affective factors that can be used to predict actual English learning outcome. Our task in the current study is different from the work by Karamouzis and Vrettos [12], in that we want to estimate student ability in English learning based on affective factors. As is commonly understood, success in education is believed to be based most on the role of cognitive factors. While this may sound acceptable conceptually, empirical research on methods for estimating student English ability using their affective parameters has not yet been fully explored.

mains ideally are considered when a study on their roles is set to examine one students success in learning. In the area of language learning, affective factors are factors that are related to learners emotional states and their attitudes toward the target language. In affective issues there has been extensive research in which most researchers have taken the common basic understanding that affective factors play an important role in language learning. These researchers, however, have different opinions about the factors underlying affective factors. Brown [5], for example, holds the view that affective factors are those factors that come from the learners themselves. Ellis [4] meanwhile has different ideas on affective factors in second language acquisition. Ellis says that these affective factors are inuenced by personality factors, such as anxiety and that how anxiety affects learning depends on learning conditions. Classication by researchers generally falls into the following: self-esteem, inhibition, risk taking, anxiety, empathy, extroversion, and motivation. Based on the various classications of the affective factors described previously, we propose three major factors to be put under the general heading of affective factors. These are motivation, attitude, and personality. The components of each of these factors are further identied by exploring each factor conceptually. The motivation factor can be divided into integrative, instrumental, resultative, intrinsic, global, situational, and task-based. The attitude factor can be divided into attitudes toward the community, English, and learning. Personality factor can be divided into introversion, extroversion, anxiety, self-esteem, and inhibition. The features of these motivation, attitude, and personality factors are shown in Tables 13, respectively. In this study, these three major factors and their corresponding components are utilized as the basis for exploring student success in learning English. The choice of affective factors as main factors is based on the work by Immordino-Yang and Damasio [7] dening the primary role of affective factors in learning.

4. English Abilities Estimation Model

4.1. Data Collection on Students Affective Factors and Instrument Validation
A questionnaire is used to quantify students affective factors as knowledge suitable for NN input. The questionnaire was developed using Likerts ve-level response scale from 5, indicating strong agreement to 1, indicating strong disagreement. A literature review was conducted to select parameters for affective factors. We have summarized and arranged these into the three major factors mentioned in the previous section, i.e., motivation, attitude, and personality. Each factor has sub-factors reected in motivation, attitude, and personality, as shown in Fig. 1. The questionnaire is a modied version of questions in the instruments developed by Horwitz et al. [16]. The questionnaire we developed was then validated by Vol.16 No.3, 2012

3. Affective Factors
Bloom [15] classies human learning potential that can be explored within an education contexts into cognitive, affective, and psychomotor domains. These three do376

Journal of Advanced Computational Intelligence and Intelligent Informatics