Diverse Ensemble Framework (DEF) for Predictive Analytics

Abstract

Machine Learning technique has numerous benefits that include high flexibility and power, lack of parametric assumption etc. Building an effective Machine Learning ensemble model, by using different categories models to perform ensembling than the concept of diversity in ensemble has occurs. To perform ensembling on the datasets the technique like bagging, boosting, and voting are used. In this thesis we take the two datasets of regression data and initially execute both the datasets on eight different machine learning models (RF, NN, LM, Cubist, Enet, LR with SS, PCR and ICR), after performing ensemble techniques on the following machine learning model we got the batter result as compare to individual results. In classification problems, observations fall into pre-assigned groups. Examples include identifying customers who would buy a product, and detecting whether a credit card expense is made by a customer. A popular approach to tackle these problems is using a collection of models that combines the collective knowledge of them. It has been shown that employing multiple models outperforms a single model. A common approach has been to use the same collection for all observations, which is also known as the static approach. Recently, there have been more attempts in using a different collection that is more specialized for each observation, depending on the features of observations. In the ensemble result seen that the ensemble of two weak models (with minimum accuracy) are gives the best results (with high accuracy) in the both the case of datasets. To make the work more efficient we proposed a framework known as Diversity framework (DEF). The proposed framework is also predicting the odor of the chemicals so it is known as olfaction prediction. The framework DEF is developed using R language and various R packages. The performance of DEF is evaluated and results show that DE framework out performs than other existing techniques like Bagging, Boosting etc.