Figures (6)  Tables (5)

    • Figure 1. 

      Research methodology used.

    • Figure 2. 

      Seeders used: (a) Sulky-Burel S.A., and (b) Nardi-TSD.

    • Figure 3. 

      Compound stacking model.

    • Figure 4. 

      Effect of basic operating parameters on HI: (a) seeder type, (b) sowing depth, and (c) sowing speed.

    • Figure 5. 

      SHAP diagram showing the relative importance and direction of influence of operators on the prediction of the HI.

    • Figure 6. 

      Feature importance: seeder parameters' impact on HI.

    • Properties Nardi-TSD Sulky-Burel S.A.
      Mounting type Trailed type Mounted type
      Total seeder width 452 cm 390 cm
      Effective working width 300 cm 300 cm
      Tire specifications 24–14.9 7.5–14
      Hopper capacity Dual hoppers: seed (540 kg) and fertilizer (550 kg) Single hopper: seed (250 kg)
      Lifting/lowering mechanism Hydraulic cylinder Manual hand lever
      Number of the furrow opener 20 20
      Row spacing 15 cm 15 cm
      Furrow opener type Shovel type Hoe Type
      Penetration angle Severe, less than 90° Obtuse, greater than 90°
      Opener arrangement Two-row, staggered (alternating) Single row
      Covering mechanism Rear leveling tines (covering tines) Rear plastic flaps and oscillating (pulsating) leveling tines
      Metering mechanism Fluted roller Studded roller
      Seed delivery tubes Helicoidally tubes Telescopic seed tube

      Table 1. 

      Technical properties of the 'Nardi-TSD' and 'Sulky-Burel S.A.' seeders used.

    • Model Hyperparameters Search range (GridSearchCV) Optimal value
      Linear regression Parameters – Default scikit-learn settings (OLS)
      RF n_estimators [50, 100, 200, 300, 500] 300
      max_depth [5, 10, 20, None] 20
      min_samples_split – 2
      SVR Kernel ['linear', 'rbf'] 'rbf'
      C [0.01, 0.1, 1, 10, 100] 10
      Gamma – Scale
      Stacking Final estimator – Linear Regression
      Base estimators - RF, SVR
      Tuning Cross-Validation – GroupKFold (n_splits = 3)

      Table 2. 

      Hyperparameter settings for the machine learning models.

    • Characteristics N Std. deviation Shapiro-Wilk test Levene's test Kruskal-Wallis test
      W p-Value p-Value H p-Value
      Seeder type Sulky 48 3.48 0.85 0.000 0.000 1.12 0.28
      Nardi 48 4.12 0.9 0.001
      Sowing depth (cm) 2.5 24 1.42 0.98 0.89 0.00 78.53 0.00
      5 24 1.42 0.98 0.98
      7.5 24 1.58 0.9 0.03
      10 24 2.01 0.92 0.07
      Sowing speed (km/h) 4.39 32 3.66 0.95 0.25 0.007 0.74 0.68
      6.42 32 3.58 0.95 0.2
      8.18 32 4.26 0.89 0.004

      Table 3. 

      Statistical analysis and non-parametric difference test.

    • InteractionsSig.
      Seeder type × sowing depth0.00
      Seeder type × sowing speed0.21
      Sowing depth × sowing speed0.00
      Seeder type × sowing depth × sowing speed0.00

      Table 4. 

      GZLM test results.

    • MAPE nRMSE MSE MAE R2 Models
      Training 0.04 0.1 2.23 1.19 84.4 Linear regression
      0.01 0.02 0.16 0.3 98.91 RF
      0.00 0.03 0.21 0.29 98.51 SVR
      0.00 0.03 0.19 0.28 98.61 Stacking
      Testing 0.06 0.2 5.55 2.04 62.2 Linear regression
      0.05 0.16 3.66 1.35 75.08 RF
      0.06 0.2 5.41 1.83 63.19 SVR
      0.06 0.19 5.12 1.77 65.13 Stacking

      Table 5. 

      Evaluating the efficiency of the models and comparing performance.