@article{fdi:010093436,
  title = {{P}roper account of auto-correlations improves decoding performances of state-space (semi) {M}arkov models},
  author = {{B}ez, {N}icolas and {G}loaguen, {P}. and {E}tienne, {M}. {P}. and {L}anco, {S}. and {J}oo, {R}. and {R}ivot, {E}. and {W}alker, {E}. and {W}oillez, {M}. and {M}ah{\'e}vas, {S}.},
  editor = {},
  language = {{ENG}},
  abstract = {{S}tate-space models are widely used in ecology to infer hidden behaviors. {T}his study develops an extensive numerical simulation-estimation experiment to evaluate the state decoding accuracy of four simple state-space models. {T}hese models are obtained by combining different {M}arkovian specifications ({M}arkov and semi-{M}arkov) for the hidden layer with the absence (model {AR}0) and presence ({AR}1) of auto-correlation for the observation layer. {M}odel parameters are issued from two sets of real annotated trajectories. {T}hree metrics are developed to help interpret model performance. {T}he first is the {H}ellinger distance between {M}arkov and semi-{M}arkov sojourn time probability distributions. {T}he second is sensitive to the overlap between the probability density functions of state-dependent variables (e.g., speed variables). {T}he third quantifies the deterioration of the inference conditions between {AR}0 and {AR}1 formulations. {I}t emerges that the most sensitive model choice concerns the auto-correlation of the random processes describing the state-dependent variables. {O}pting for the absence of auto-correlation in the model while the state-dependent variables are actually auto-correlated, is detrimental to state decoding performance. {R}egarding the hidden layer, imposing a {M}arkov structure while the state process is semi-{M}arkov (with negative {B}inomial sojourn times) does not impair the state decoding performances. {T}he real-life estimates are consistent with our experimental finding that performance deteriorates when there are significant temporal correlations that are not accounted for in the model. {I}n light of these findings, we recommend that researchers carefully consider the structure of the statistical model they suggest and confirm its alignment with the process being modeled, especially when considering the auto-correlation of observed variables.},
  keywords = {},
  booktitle = {},
  journal = {{PEER} {C}ommunity {J}ournal},
  volume = {5},
  numero = {},
  pages = {e38 [35 ]},
  year = {2025},
  DOI = {10.24072/pcjournal.535},
  URL = {https://www.documentation.ird.fr/hor/fdi:010093436},
}