Learning of an Instance-Based System for Predicting Garment Sizes

Abstract

This project developed an instance-based system to predict garment sizes. Instance-based systems learn by increasing the size of the instance base. Before installing a system, developers need to know how many instances is enough to achieve the desired decision accuracy. Including more instances than necessary can be wasteful of development time and wastes storage capacity. Knowing the learning curve could be helpful in deciding how many instances is enough. Among the few reported learning curves for instance-based systems, most common is the classic learning curve, a rapid increase in accuracy as the first instances are added, then a tapering off as the number of instances increases. Therefore, it appears sensible for most systems to add enough instances until the curve levels off, but no more. In this paper we review the learning phenomenon in natural and artificial systems, summarize the learning behavior of several instance-based systems, and describe how we determined the learning curve of the size prediction system.