A comparative study of classifier techniques for lift index data analysis

Purpose A classifier technique is one of the important tools which may be used to classify the data or information into systematic manner based on certain criteria pertaining to get the accurate statistical information for decision making. It plays a vital role in the various applications, such as business organization, e-commerce, health care, scientific and engineering application. The purpose of this paper is to examine the performance of different classification techniques in lift index (LI) data classification. Design/methodology/approach The analyses consist of two stages. First, the random data are generated for lifting task through computer programming, which is then put into the National Institute for Occupational Safety and Health equation for LI estimation. Based on the evaluated index, the task may be classified into two groups, i.e. high-risk and low-risk task. The classified task is considered to analyze the performance of different tools like Artificial Neural Network (ANN), discriminant analysis (DA) and support vector machines (SVMs). Findings The work clearly demonstrates the accuracy and computational ability of ANN, DA and SVM for data classification problems in general and LI data in particular. From the research it may be concluded that SVM may outperform ANN and DA. Research limitations/implications The research is limited to a particular kind of data that may be further explored by selecting the different controllable parameters and model specification. The study can also be applied to realistic problem of manual loading. It is expected that this will help researchers, designers and practicing engineers by making them aware of the performance of classification techniques in this area. Originality/value The objective of this research work is to assess and compare the relative performance of some well-known classification techniques like DA, ANN and SVM, which suggest that data characteristics considerably impact the classification performance of the methods.