Development of a model for detecting diseases of crops using spectral and vegetative features
Abstract
This study presents a leakage-aware framework for feature engineering and spatial data quality assessment of multitemporal Sentinel-2 observations for agricultural land classification. After removing 29 duplicate coordinates, 5,563 samples formed 1,983 connected components under a 5 km rule and were assigned to spatially independent training, validation, and held-out test subsets. Twelve spectral bands and eight vegetation and moisture indices generated 648 spectral, 432 index-derived, and 1,080 combined features. On validation data, the combined representation achieved a macro-F1 of 0.7261 with HistGradientBoosting, while the proposed out-of-fold (OOF)-stacked multispectral machine learning fusion (OS-MMF) achieved 0.7408. Across ten repeated whole-component partitions, the combined representation obtained a mean macro-F1 of 0.7362. Removing the redundant McFeeters normalized difference water index (NDWI) features produced comparable performance. Final held-out evaluation of OS-MMF yielded an accuracy of 0.7803 and a macro-F1 of 0.7080, supporting reproducible spatially controlled classification without establishing universal model superiority.
Keywords
Agricultural land classification; Machine learning; Precision agriculture; Spectral analysis; Vegetation indices
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.12164
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.