1 Copyright © 2011 by ASME Proceedings of the 2011 ASME International Mechanical Engineering Congress & Exposition IMECE2011 November 11-17, 2011, Denver, Colorado IMECE2011-63409 ARTIFICIAL NEURAL NETWORKS FOR PREDICTING THE PERFORMANCE OF NOVICE CAD USERS BASED ON THEIR PROFILED TECHNICAL ATTRIBUTES R. F. Hamade American University of Beirut Mechanical Engineering Department P.O. Box 11-0236, Riad El-Solh Beirut 1107 2020, Lebanon Email: rh13@aub.edu.lb A. H. Ammouri American University of Beirut Mechanical Engineering Department P.O. Box 11-0236, Riad El-Solh Beirut 1107 2020, Lebanon Email: ahe11@aub.edu.lb H. A. Artail American University of Beirut Department of Electrical and Computer Engineering P.O. Box 11-0236, Riad El-Solh Beirut 1107 2020, Lebanon Email: ha27@aub.edu.lb ABSTRACT This paper utilizes Artificial Neural Networks (ANN) to forecast the mechanical CAD performance of novice trainees involved in formal training. We utilize 3 Artificial Neural Networks, ANN, techniques: Feed-Forward Backpropagation, Elman Backpropagation, and Generalized Regression. We also compare their predictive capabilities compared to those of linear regression techniques. For this purpose, two kinds of data are utilized as input vectors for the predictive techniques: performance data and trainee attributes data. Such data has been previously published by Hamade and coworkers. Performance data is based on the following four quantitative measures of performance: (1) construction speed of the CAD model, (2) sophistication of the constructed CAD model, and the rates of improvement of (3) construction speed and (4) model sophistication. Trainees‟ attributes identified as related to building CAD competence include: (1) technical and (2) character attributes and (3) learning styles. Strong correlations have been found between many of the trainees‟ profiled attributes and trainee‟s actual performance throughout and upon the conclusion of the training. Generally, the ANN methods as well as the linear regression techniques were found to be successful in discriminating the trainees based on their profiled attributes. However, the findings suggest that, of the networks considered, the Generalized Regression Neural Network (GRNN) gave the best prediction results by yielding the least prediction error practically across all performance measures. Therefore, GRNN can be used to predict the performance of the novice CAD users. This capability may be used to pre-assess the development of CAD skills as training progresses and may serve as basis to develop custom CAD training programs and to improve the efficiency and effectiveness of CAD training. 1. INTRODUCTION An important aspect of developing sustainable CAD expertise at the corporate level is in line with recent trends that stress the accomplishment of sustainable corporate productivity [1]. Consequently, organizations and educational institutions alike are increasingly seeking innovative ways to improve the effectiveness of CAD training [2-3]. In turn, the development of individual CAD expertise depends on the effective and efficient knowledge acquisition by the trainees‟ so that they achieve CAD competence [4]. Recent work by Hamade and coworkers [5] has investigated the technical attributes that may affect the trainees‟ ability to acquire CAD knowledge as they underwent a formal training course in a university setting. Specifically, trainees‟ answers to written questionnaires were used to quantify these attributes. Furthermore, CAD performance assessment was accomplished by using measures related to the speed and the sophistication of constructed CAD models. This involved constructing a performance time (speed, or simply, T) learning curve and a sophistication learning based on the number of features used to build test models (or simply, F) curve for each trainee as master learning curves for the whole class. The trainees‟ profiled technical attributes included such topics as basic math, advanced, math, mechanical design