A Novel EN-TLBO-SVR Model for Analyzing Achievements of Government Schemes

Abstract

Forecasting physical achievements of welfare and developmental schemes offered by government is always a challenging assignment. Studies undertaken to address this problem using data mining techniques are not fully efficient when the number of features is more than that of samples. This paper presents a novel hybrid machine learning model based on support vector regression, which exhibits magnificent generalisation capability on small samples with proper selection of hyper-parameters using teaching-learning-based optimisation along with elastic net for feature selection. For predicting achievement, a dataset pertaining to housing scheme of Government of India is used where number of samples available is small. It is observed that the proposed hybrid model EN-TLBO-SVR has not only outperformed the use of particle swarm optimisation for hyper-parameter selection, but also kernel principal component analysis and sequential forward floating selection for dimensionality reduction with an additional advantage of identifying significant features present in the samples.