@article{fdi:010096465,
  title = {{M}achine-learning crop-type mapping sensitivity to feature selection and hyperparameter tuning},
  author = {{P}erez-{F}lores, {M}. and {S}atg{\'e}, {F}r{\'e}d{\'e}ric and {M}olina-{C}arpio, {J}. and {H}ostache, {R}enaud and {P}illco-{Z}olá, {R}. and {T}ola, {D}. and {U}scamayta-{F}errano, {E}. and {B}ustillos, {L}. and {B}onnet, {M}arie-{P}aule and {D}uwig, {C}{\'e}line},
  editor = {},
  language = {{ENG}},
  abstract = {{H}ighlights {W}hat are the main findings? {C}rop-type mapping reliability is highly sensitive to features selection and hyperparameter tuning. {T}he model and target independent {VIF} feature selection is not recommended for crop-type mapping {W}hat are the implications of the main findings? {M}ost reliable crop-type mapping is obtained through a proposed three-step process combining wrapped features selection with hyperparameter tuning. {B}ased on open-access data and software, the proposed method can be used to support agriculture monitoring in a complex socio-economic context. {H}ighlights {W}hat are the main findings? {C}rop-type mapping reliability is highly sensitive to features selection and hyperparameter tuning. {T}he model and target independent {VIF} feature selection is not recommended for crop-type mapping {W}hat are the implications of the main findings? {M}ost reliable crop-type mapping is obtained through a proposed three-step process combining wrapped features selection with hyperparameter tuning. {B}ased on open-access data and software, the proposed method can be used to support agriculture monitoring in a complex socio-economic context.{A}bstract {T}o improve crop yields and incomes, farmers consistently adapt their practices to climate and market fluctuations, resulting in highly variable crop field distribution and coverage in space and time. {A}s these dynamics illustrate farmers' challenges, up-to-date crop-type mapping is essential for understanding farmers' needs and supporting their adoption of sustainable practices. {W}ith global coverage and frequent temporal observations, remote sensing data are generally integrated into machine learning models to monitor crop dynamics. {U}nlike physical-based models that rely on straightforward use, implementing machine learning models requires extensive user interaction. {I}n this context, this study assesses how sensitive the models' outputs are to feature selection and hyperparameter tuning, as both processes rely on user judgment. {T}o achieve this, {S}entinel-1 ({S}1) and {S}entinel-2 ({S}2) features are integrated into five distinct models ({R}andom {F}orest ({RF}), {S}upport {V}ector {M}achine ({SVM}), {L}ight {G}radient {B}oosting ({LGB}), {H}istogram-based {G}radient {B}oosting ({HGB}), and {E}xtreme {G}radient {B}oosting ({XGB})), considering several features selection ({V}ariance {I}nflation {F}actor ({VIF}) and {S}equential {F}eature {S}elector ({SFS})) and hyperparameter tuning ({G}rid-{S}earch) setup. {R}esults show that the preprocess modeling feature selection ({VIF}) discards the features that the wrapped method ({SFS}) keeps, resulting in less reliable crop-type mapping. {A}dditionally, hyperparameter tuning appears to be sensitive to the input features, and considering it after any feature selection improved the crop-type mapping. {I}n this context a three-step nested modeling setup, including first hyperparameter tuning, followed by a wrapped feature selection ({SFS}) and additional hyperparameter tuning, leads to the most reliable model outputs. {F}or the study region, {LGB} and {XGB} ({SVM}) are the most (least) suitable models for crop-type mapping, and model reliability improves when integrating {S}1 and {S}2 features rather than considering {S}1 or {S}2 alone. {F}inally, crop-type maps are derived across different regions and time periods to highlight the benefits of the proposed method for monitoring crop dynamics in space and time.},
  keywords = {crop-type mapping ; sentinel ; machine learning ; {B}olivia ; {A}ltiplano ; {BOLIVIE}},
  booktitle = {},
  journal = {{R}emote {S}ensing},
  volume = {18},
  numero = {4},
  pages = {563 [23 p.]},
  ISSN = {2072-4292},
  year = {2026},
  DOI = {10.3390/rs18040563},
  URL = {https://www.documentation.ird.fr/hor/fdi:010096465},
}