A Modified Neurofuzzy Based Quality of eLearning Model (Modified SCeLQM)

Higher Education Institutions are required to invent means and tools to meet the increasing demand of students’ enrolment. ELearning is one of these potential means which indicates that the issue of the quality of eLearning models is essential. A developed two-stage, Multi-Input-Single-Output SCeQLM a eLearning quality model has been reviewed. SCeLQM is based on ten Critical Success Factors, CSF. Each CSF consists of several characteristics or sub-factors. Stage one models, individually, every CSF using the rule-based soft Computing, Neurofuzzy in particular, approach, where relative sub-factors are input into the models. Stage 2 feeds the outputs from the processed ten CSFs models, with equal weights, into another Neurofuzzy-based model to produce a unique value that describes the status of the quality of the eLearning system in the higher education institution under consideration. The output of SCeLQM will be one of the categories POOR, FAIR, GOOD, V. GOOD and EXCELLENT, in the ranges 1.0 to >2.3, 2.3 to <3.2, 3.2 to < 4, 4 to <4.5 and 4.5 to 5, respectively. Several metrics have been used to measure the adequacy of the SCeLQM model. 338 data sets were divided into 80% training and 20% checking data sets using the cross validation approach. The obtained consistent and promising results of these metrics, above 0.99 for Correlation Coefficient and below 1.722 for the Mean Absolute Percentage Error, suggest the suitability to apply the modeling techniques, Neurofuzzy, in this type of problems. This paper focuses on the weights of the ten inputs, second stage, and their impact on the overall output of the SCeLQM model. Weights of one input, PEDAGOGY CSF, have been doubled several times to obtain weights of twice, four times, eight times, sixteen times the equal weights of all other nine inputs. Similarly, the available 338 data sets have been cross validated into 80% training data sets and 20% data sets. Four measures have been used to validate and check the proposed models. These metrics include the Correlation Coefficient, CC, the Mean Absolute Percentage Error, MAPE, the Maximum Difference, MD, and the Maximum Difference Percentage, MDP. The achieved CC range between 0.999 and 0.908, MAPE values vary between 0.1382 and 6.625, MD values range a SCeLQM model has been recently submitted for publication entitled “A Neurofuzzy-based Quality of eLearning Model” between 0.061 and 0.8 and MDP values range between 1.22 and 16 for the various models. A comparison study shows that the four measures follow quadratic trend, different parameters, with the weights’ variations. It is found that a 2% threshold of CC values (0.02 below the optimum value of one) yields significant changes of the overall output that corresponds to greater than or equal to the 30.8% weight case. Regarding the MAPE metric, it is found that an increase of four MAPE value thresholds (4.0 above the optimum value of zero) will produce significant changes to the overall output that are obtained at greater than or equal to the 47.1% weight cases. A rise of 0.5 MD value thresholds (0.5 above the optimum value of zero) will make significant changes of the overall output at greater or equal the 47.1% weight cases. Whereas; an increase of 8 MDP value thresholds (8.0% above the optimum value of zero) will produce significant changes of the overall output are obtained at greater or equal the 47.1% weight cases. Furtherlly, the five categories, POOR, FAIR, GOOD, V. GOOD and EXCELLENT, of the SCeLQM overall output has been addressed. It is found that these categories will be affected, either improving or worsening, when the one of the weight of an input has been set to equal or higher than four times than the equal weights of the other nine inputs. This value corresponds to 0.984 correlation coefficient, 2.437 MAPE, 0.31 MD, and 6.2 MDP values and four times (30.8%) weight of one input of the equal weight of the other nine inputs. That is, considerable contributions of the weight of one input will affect the overall model output when it is higher than four times. Additionally, the variations of CC values against the number of categories’ changes follow a second order quadratic trend. The achieved promising, consistent and promising results of these metrics suggest the suitability to apply the modeling techniques, Neurofuzzy, in this type of problems. It is intended to further investigate, enhance and address the impact of the rules and develop a Web-based version of SCeLQM model in the near future. Keywords--eLearning, Quality eLearning Models, Quality, Higher Education, Soft Computing. International Journal of Computer and Information Technology (ISSN: 2279 – 0764) Volume 03 – Issue 06, November 2014 www.ijcit.com 1328