Locally weighted regression with different kernel smoothers for software effort estimation

Estimating software effort has been a largely unsolved problem for decades. One of the main reasons that hinders building accurate estimation models is often heterogeneous nature data with complex structure. Typically, from local tend to be more than using entire data. Previous studies have focused on use clustering techniques and decision trees generate coherent can help in prediction models. However, these approaches may fall short some aspect due limitations finding optimal clusters processing noisy In this paper we used sophisticated locality approach mitigate shortcomings Locally Weighted Regression (LWR). This method provides an efficient solution learn by model combines multiple regression k-nearest-neighbor based model. The factor affecting accuracy choice kernel function derive weights investigates effects choosing different kernels performance problem. After comprehensive experiments 7 datasets, 10 kernels, 3 polynomial degrees 4 bandwidth values total 840 variants, found that: 1) Uniform functions cannot outperform non-uniform functions, 2) type, parameters no specific effect accuracy. other words, change or degree occurred significant difference rankings. short, methods Triweight Triangle perform better kernels. Hence, encourage as smoother wide small degree.