Radar Data Tables

Scope and purpose These tables provide the inputs and normalized scores used to generate the normalized radar charts that compare methods M1, M2, M3, Proposed across datasets (Corn (control), Chinese Kale, Rock Melon, Tomato Ambient, Tomato S1, Tomato S2, Tomato S3–S6, Tomato S7). They enable third parties to reproduce the figures and verify trends without access to raw time-series. Files Radar_Raw.csv — raw (unnormalized) metrics per dataset × method. Radar_Normalized.csv — per-dataset 0–1 scores for radar axes (higher = better). Table_B_Efficiency.csv — runtime/resource context (referenced by efficiency figures). File schema 1) Radar_Raw.csv Column Meaning Units/Notes Dataset Dataset name (e.g., “Tomato S3–S6”) String Method One of {M1, M2, M3, Proposed} String MAPE Mean Absolute Percentage Error % (numeric) RMSE Root Mean Squared Error metric units of variable; numeric VarReduction_pct Variance reduction vs. raw % (higher is better) OutlierSuppression_pct Suppressed outliers vs. raw % (higher is better) Latency_seconds Processing time per sample/segment (see paper) seconds; numeric Estimated “No” if value is reported explicitly in the manuscript; “Yes” if logically estimated to avoid blanks in comparisons String {Yes, No} Notes: The file preserves unusual values as reported (e.g., negative RMSE printed for one case), to reflect the manuscript faithfully. 2) Radar_Normalized.csv Column Meaning Notes Dataset Dataset name String Method {M1, M2, M3, Proposed} String Accuracy_from_RMSE Normalized score (higher = better) from RMSE RMSE inverted + min–max normalized per dataset Error_from_MAPE Normalized score (higher = better) from MAPE MAPE inverted + min–max normalized per dataset Robustness Normalized OutlierSuppression_pct min–max normalized per dataset Smoothness Normalized VarReduction_pct min–max normalized per dataset Latency Normalized (higher = better) from latency Latency inverted + min–max normalized per dataset Normalization (per dataset) To make all axes “higher = better,” we apply min–max scaling within each dataset: For metrics where higher is better (Robustness, Smoothness):score = (x − min) / (max − min) For metrics where lower is better (RMSE, MAPE, Latency):score = (max − x) / (max − min) This yields values in [0, 1] for each axis and avoids cross-dataset scale confounds. The radar axes are then: Accuracy_from_RMSE (from RMSE, inverted) Error_from_MAPE (from MAPE, inverted) Robustness (OutlierSuppression_pct) Smoothness (VarReduction_pct) Latency (from Latency_seconds, inverted) Reproducing the radar charts Filter Radar_Normalized.csv to one Dataset. Pivot to a 5×4 grid (rows = axes above, columns = methods). Plot a radar/spider chart with values in [0,1]. (Any software is fine: Python/matplotlib, R, Excel.) Optional: annotate with the corresponding raw values by joining to Radar_Raw.csv for context. Reproducibility & provenance Any cell marked Estimated = “Yes” in Radar_Raw.csv is a logical estimate added solely to avoid blanks and to enable like-for-like plotting across all methods/datasets. Exact values from the manuscript are marked Estimated = “No”. Radar_Normalized.csv is derived from Radar_Raw.csv using the rules above; no further smoothing or weighting is applied. Efficiency summaries used in the paper’s runtime figures are provided in Table_B_Efficiency.csv for transparency. License & citation (suggested) License: CC BY 4.0 (or your journal’s required license). How to cite (example):Supplementary Radar Data Tables for “Advancing Agricultural Time-Series Data Analysis,” Zenodo, DOI: <10.5281/zenodo.17256230>. Quick sanity checks (recommended) Within each dataset, normalized values should lie in [0, 1] and at least one method should attain 0 and one 1 on each axis. Visual ranking on the radar should be consistent with the RMSE/MAPE directionality and the Latency advantage patterns discussed in Section F.