ELECTRE-Tree

1. Draft: https://arxiv.org/abs/2007.10047
2. Article: https://www.emerald.com/insight/content/doi/10.1108/DTA-10-2020-0256/full/html
3. Try it in Colab: ( Colab Demo )

ELECTRE-Tree Algorithm to infer the ELECTRE Tri-B method parameters. The function returns: 1) A list of optimized sub-models that can be used to vote the allocation of alternatives (assign to a class) or can infer the ELECTRE Tri-B parameters using the average.

"tree_electre_tri_b" arguments

• dataset = A numpy array where the rows are the alternatives and columns are the criteria.

• target_assignment = Optional argument. A list of previous allocation (labels) of alternatives that the algorithm will try to follow (classification problem). The default value is [].

• W = Optional argument. A list of weights for each criterion indicated by the decision maker. The default value is [], meaning that the algorithm will try to optimize this parameter.

• Q = Optional argument. The indifference threshold list indicated by the decision maker. The default value is [], meaning that the algorithm will try to optimize this parameter.

• P = Optional argument. The preference threshold list indicated by the decision maker. The default value is [], meaning that the algorithm will try to optimize this parameter.

• V = Optional argument. The veto threshold list indicated by the decision maker. The default value is [], meaning that the algorithm will try to optimize this parameter.

• cut_level = Optional argument. The list of possibles cut level values indicated by the decision maker. The default value is [0.5, 1.0], meaning that the algorithm will try to optimize this parameter with a value from 0.5 to 1.

• rule = Decides if the allocation rule is pessimist 'pc' or optimist 'oc'. The default values is 'pc'.

• number_of_classes = An integer that indicate the total number of classes of the problem. The default value is 2.

• elite = The quantity of best indivduals to be preserved in the genetic algorithm. The quantity should be low to avoid being traped in local otima. The default value is 1.

• mutation_rate = Chance to occur a mutation operation in the genetic algorithm. The default value is 0.01

• eta = Value of the mutation operator used in the genetic algorithm. The default value is 1.

• mu = Value of the breed operator used in the genetic algorithm. The default value is 2.

• population_size = The population size used in the genetic algorithm. The default value is 15.

• generations = The total number of iterations used in the genetic algorithm. The default value is 150.

• samples = The percentage of the number of alternatives (randomly selected) used in each submodel. The default value is 0.10.

• number_of_models = The total number of generated sub-models. The defaul value is 100.

"predict" arguments

• models = A list of optimized sub-models generated by the "tree_electre_tri_b" function.

• dataset = A numpy array where the rows are the alternatives and columns are the criteria.

• verbose = Prints the prediction for each alternative. The default value is True.

• rule = Decides if the allocation rule is pessimist 'pc' or optimist 'oc'. The default values is 'pc'.

"metrics" arguments. Returns the inferred parameters.

• models = A list of optimized sub-models generated by the "tree_electre_tri_b" function.

• number_of_classes = An integer that indicate the total number of classes of the problem. The default value is 2.

"plot_decision_boundaries" arguments

• data = A numpy array where the rows are the alternatives and columns are the criteria.

• models = A list of optimized sub-models generated by the "tree_electre_tri_b" function.

Other MCDA Methods:

• pyDecision - A library for many MCDA methods
• pyMissingAHP - A Method to Infer AHP Missing Pairwise Comparisons
• 3MOAHP - Inconsistency Reduction Technique for AHP and Fuzzy-AHP Methods
• Ranking-Trees - Algorithm to infer the ELECTRE II, III, IV and PROMETHEE I, II, III, IV method parameters