Conservative Generalisation for Small Data Analytics -An Extended Lattice Machine Approach

Shuangshuang Kong, Hui Wang, Kaijun Wang

Abstract: Small data analytics is to tackle the data analysis challenges such as overfitting when the data set is small. There are different approaches to small data analytics, including knowledge-based learning, but most of these approaches need experience to use. In this paper we consider another approach, lattice machine. Lattice machine is a conservative generalisation based learning algorithm. It is a learning paradigm that "learns" by generalising data in a consistent, conservative and parsimonious way. A lattice machine model built from a dataset is a set of hyper tuples that tightly "wraps around" clusters of data, each of which is a conservative generalisation of the underlying cluster. A key feature of lattice machine, indeed any conservative generalisation based learning algorithm, is that it has high precision and low recall, limiting its applications as high recall is needed in some applications such as disease (e.g. covid-19) screening. It is thus necessary to improve lattice machine's recall whilst retaining his high precision. In this paper, we present a study on how to achieve this for lattice machine.