Personalizing Group Instruction Using Knowledge Space Theory and Clustering Techniques

Abstract

In the competitive market today, availability of appropriate skills and competencies that are aligned with an organization's objectives and that contribute to the organization's long-term success and survival is not always guaranteed. Lack of such alignment leads to a Skill Gap. A Skill Gap is the gap between an organization's current capabilities and the skills the employees must have to achieve an organization's goals. This is especially true for engineering companies where the half-life of knowledge is relatively short. One cause of Skill Gap is inefficient and standardized training programs. Due to high costs, most training programs are not personalized, and deliver generic training to employees which may not be very effective in addressing individual or group Skill Gaps. This thesis explores personalizing and optimizing the content delivery decisions made by workforce trainers and instructors. The proposed approach is data driven and combines a set-theoretic framework called the Knowledge Space Theory (KST) with analytic techniques like cluster analysis. In specific, K-Means, DBSCAN and EM clustering techniques are used in conjunction with KST to cluster learners based on currently acquired skills, and on skills they are ready to acquire next. These clusters of learners can be used to design personalized training/instructional programs. Various internal measures like Compactness, Separation, Dunn Index, and Davies-Bouldin Index, and external measures like Purity, Entropy, Normalized Mutual Information, and Adjusted Random Index are used to compare alternative clustering techniques. Sensitivity analysis was also carried out. In general, K-Means seems to perform better than DBSCAN and EM for this type of data. However, there is no systematic preference between prior learning as opposed to affordance for future learning to cluster data.