Class partition approach for software effort estimation using support vector machine

Software effort estimation consists of those procedures and activities which help to predict most accurate development effort as well as cost of a software product. After analyzing various proposed concept and theories regarding this we tried to give a new concept which works over partition of a data set. The partition procedure depends over the correlation of input features as well as output features. This algorithm is a generic concept which can be useful in other fields also. Previously proposed models and theories depend on the detection and prediction methodologies. In detection process, developers try to find out loopholes in the prior proposed models while in prediction model they try to develop a new theory that mitigates those loopholes. In our proposed approach we work over a mathematical model that can be a replacement of a well-known data mining algorithm K-NN (K nearest neighbor), which uses clusters for classification. Our proposed methodology is based on feature selection. This feature selection process is based on correlation coefficient that can be defined by the user. After feature selection, we tried to achieve the rate of change of the target attribute with respect to the other attributes. We have classified the data vectors according to the rate of change of target value (Effort) and attributes obtained after feature selection. Target value (Effort) will be fixing for whole processing. Finally we applied regression to obtain a mathematical model for the corresponding class.