Demand-Driven Associative Classification

Abstract

The ultimate goal of classification algorithms is to achieve the best possible classification performance for the problem at hand. Most often, classification performance is obtained by assessing some accuracy criterion using the test set, $T.$ Therefore, an effective classification in $T.$ As discussed, it is often hard to approximate the target function defined over inputs in $T,$ using a single mapping function. The key insight is to produce a specifically designed function, $f^{x_i}_{S},$ which approximates the target function at each input $x_i\in T.$ Thus, a natural way to improve classification in $T,$ on a demand-driven basis. In this case, particular characteristics of each input in $T$ may be taken into account while predicting the corresponding output. The expected result is a set of multiple mapping functions, where each function $f^{x_i}_{S}$ is likely to perform particularly accurate predictions for input $x_i\in{T},$ no matter the (possible poor) performance in predicting outputs for other inputs .