I dunno, N=66 training and N=34 for validation along with Lasso regression suggests the performance measures are dramatically overfit

See figure 3 etc: https://pubs.acs.org/doi/pdf/10.1021/acsami.1c01754