In a recent study, the researchers introduce the influence score or ‘I-score’ as a static correlated with how much variables inherently can predict that can consequently be used to identify exceedingly predictive variables.
“In our past studies, we revealed that significant variables may not necessarily be predictive and that good predictors might not appear statistically significant,” says principal investigator Shaw-Hwa Lo, a lecturer of statistics at Columbia University. ‘It left us with a vital question of how we can identify exceedingly predictive variables then, if not through a guideline of statistical significance.
“In this study, we offer a theoretical framework from which to design good measures of prediction in general. Vitally, we introduce a variable set’s prediction as a novel parameter of interest to estimate and offer the I-score as a candidate statistic to measure the variable set prediction,” says the head author of the research.
Present approaches to prediction usually comprise significance – based parameter for estimating variables to utilize in models evaluating models and variables simultaneously for prediction utilizing cross – validation or independent test data.
“Utilizing the I-score prediction framework enables us to define a novel estimate of predictively based on the analysed data that will in turn allows assessing variable sets for, high prediction,” says Lo. It was also added that while intuitive obvious, not ample attention has been levied to the consideration of prediction as a parameter of interest to measure.
In this study, the researchers illustrate the prediction for a variable set and reveal that a simple sample measurement of prediction directly does not offer usable information for the prediction oriented researcher. They go on to illustrate that the I-score can be utilized to compute a measure that asymptotically approaches prediction. I-score can efficiently differentiate between predictive and noisy variables, explains Lo, making it helpful in variable selection.
A further advantage is that while usual methods need heavy use of cross validation information or testing data to analyse the predictors, the I-score method does not depend as much on this. “We provide simulations and an application of the I-score on real information to illustrate the statistic’s predictive performance on sample data,” he says. “It shows that the I-score can gather exceedingly predictive variable sets estimate a lower bound for the theoretical accurate prediction rate and correlates well with the out. We suggest that utilizing such I-score technique can support in finding variables sets with lucrative prediction rates, but, further studies in the avenue of sample – based estimated of prediction is required.”
‘We are expecting to impress upon the scientific department the notion that for those who may not be interested in predicting the outcome of interest, must with rather intricate or high dimensional information, we might gain be reconsidering the question as how to search for exceedingly predictive variables,” says Lo
Filed Under: News