Abstract
In knowledge-based systems, besides obtaining good output prediction accuracy, it is crucial to understand the subset of input variables that have most influence on the output, with the goal of gaining deeper insight into the underlying process. These requirements call for logistic model estimation techniques that provide a sparse solution, i.e., where coefficients associated with non-important variables are set to zero. In this work we compare the performance of two methods: the first one is based on the well known Least Absolute Shrinkage and Selection Operator (LASSO) which involves regularization with an ℓ 1 norm; the second one is the Relevance Vector Machine (RVM) which is based on a Bayesian implementation of the linear logistic model. The two methods are extensively compared in this paper, on real and simulated datasets. Results show that, in general, the two approaches are comparable in terms of prediction performance. RVM outperforms the LASSO both in term of structure recovery (estimation of the correct non-zero model coefficients) and prediction accuracy when the dimensionality of the data tends to increase. However, LASSO shows comparable performance to RVM when the dimensionality of the data is much higher than number of samples that is p > > n .
Highlights
Techniques for the estimation of sparse models have gained increasing attention in the last two decades and have found several practical applications in different areas of science and engineering
Assuming that the output variable y is distributed in logistic regression according to a Bernoulli distribution, the probability of the outcome for the i data point can be written in compact form as: p(yi | xi ; θ) = σθ ( xi )yi (1 − σθ ( xi ))1−yi
In general, the performance for all indicators improves for both methods with increasing values of n, for example the misclassification error (MCE) in Figure 5 for dataset (f) decreases as n increases
Summary
Techniques for the estimation of sparse models have gained increasing attention in the last two decades and have found several practical applications in different areas of science and engineering. In high dimensional settings, there is a high number of measured variables and not all of them are relevant in terms of correlation with the output. In these cases, sparse models are important to avoid overfitting and improve model prediction performance as well as to identify a subset of input variables representing the most important drivers of the output variation [8]. Filter methods first identify a subset of variables, for example using correlation analysis, which is used as input with standard algorithms.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have