Improving state estimation through projection post-processing for activity recognition in football

The past decade has seen an increased interest in human activity recognition. Most commonly, the raw data coming from sensors attached to body parts are unannotated, which creates a need for fast labelling method. Part of the procedure is choosing or designing an appropriate performance measure. We propose a new performance measure, the Locally Time-Shifted Measure, which addresses the issue of timing uncertainty of state transitions in the classification result. Our main contribution is a novel post-processing method for binary activity recognition. It improves the accuracy of the classification methods, by correcting for unrealistically short activities in the estimate.

Interpretable random forest models through forward variable selection

Random forest is a popular prediction approach for handling high dimensional covariates. However, it often becomes infeasible to interpret the obtained high dimensional and non-parametric model. Aiming for obtaining an interpretable predictive model, we develop a forward variable selection method using the continuous ranked probability score (CRPS) as the loss function. Our stepwise procedure leads to a smallest set of variables that optimizes the CRPS risk by performing at each step a hypothesis test on a significant decrease in CRPS risk. We provide mathematical motivation for our method by proving that in population sense the method attains the optimal set. Additionally, we show that the test is consistent provided that the random forest estimator of a quantile function is consistent. In a simulation study, we compare the performance of our method with an existing variable selection method, for different sample sizes and different correlation strength of covariates. Our method is observed to have a much lower false positive rate. We also demonstrate an application of our method to statistical post-processing of daily maximum temperature forecasts in the Netherlands. Our method selects about 10% covariates while retaining the same predictive power.