Usage of Artificial Intelligence and Remote Sensing as Efficient Devices to Increase Agricultural System Yields
Table 3
Advantages and limitations of algorithms employed for wide-area classification of satellite picture records [147].
Algorithm
Advantages
Limitations
Maximum likelihood
(i) Easy application (ii) Simple to comprehend and interpret (iii) Forecasts category membership probability
(i) Parametric (ii) Supposes normal distribution of records (iii) Elevated training sample needed
Artificial neural networks
(i) Manages big attribute space well (ii) Shows strength of class membership (iii) Normally high classification precision (iv) Challenge to training records deficiencies—needs less training records than Decision Trees
(i) Requires factors for network modeling (ii) Tends to overfit records (iii) Black box (rules are unidentified) (iv) Computationally powerful (v) Time-consuming training
Support vector machines
(i) Manages large feature space well (ii) Insensitive to Hughes consequence (iii) Works well with little training data sets (iv) Does not overfit
(i) Requires factors: regularization and core (ii) Reduced performance with limited attribute space (iii) Computationally powerful (iv) Created as binary, even though variations are present
Decision trees
(i) No requirement for any sort of factor (ii) Simple to use and understand (iii) Handles absent records (iv) Handles records of diverse types and degrees (v) Handles nonlinear connexions (vi) Not sensitive to noise
(i) Susceptible to noise (ii) Are inclined to overfit (iii) Does not perform as well as others in big attribute spaces (iv) Big training test needed
Random forests
(i) Ability to establish variable significance (ii) Strong to data diminution (iii) Does not overfit (iv) Generates unbiased precision estimate (v) Higher precision than Decision Trees
