Creation of neuro-fuzzy expert system from online analytical processing (olap) tools

A computer-implemented method for generation of a neuro-fuzzy expert system, the computer-implemented method comprising:

• generating a neuro-fuzzy expert system from input data with a neural network structure comprising a plurality of layers including:
• a first layer with a plurality of first layer nodes with a number of first layer nodes determined by at least one clustering algorithm performed on a set of input values associated to each input attribute with one first layer node defined for at least one cluster, and correspondence established between each first layer node and one input attribute for fuzzifying by determining each membership value of input attributes, and passing each membership value as first layer outputs to a second layer;
• the second layer with a plurality of second layer nodes with each second layer node representing a unique combination of nodes in the first layer that correspond to different input attributes, and links established between each second layer node and each first layer node from a corresponding unique combination for integrating the first layer outputs and passing them to a third layer as second layer outputs;
• the third layer with a plurality of third layer nodes with a number of third layer nodes determined by at least one clustering algorithm on a set of values associated to each output attribute with one third layer node defined for at least one cluster, and correspondence established between each third layer node and one output attribute, and links established between at least one third layer node and at least one node in the second layer for applying weights to the second layer outputs and for aggregating second layer outputs and for defuzzifying the second layer outputs to produce crisp output values;
• training of a neural network of the neuro-fuzzy expert system with the input values associated to each input attribute and output values associated to each output attributes so that a weight of at least one link between the second layer and third layer is adjusted based on a signal that comes from at least one third layer node and a signal from at least one second layer node.

A computer-implemented method for automatic generation of a neuro-fuzzy expert system, the computer-implemented method comprising:

• generating a neuro-fuzzy expert system from input data with a neural network structure comprising a plurality of layers including:
• a first layer with a number of first layer nodes equal to a number of input attributes, and each first layer node receiving values of at least one input attribute and passing the values of the input attributes as first layer outputs to a second layer;
• the second layer with a plurality of second layer nodes with a number of second layer nodes determined by at least one clustering algorithm performed on a set of values associated to each input attribute with one second layer node defined for at least one cluster, and links established between each second layer node and one first layer node corresponding to that input attribute for fuzzifying by determining each membership value of first layer outputs, and passing each membership value to a third layer with links established between at least one third layer node and at least one second layer node for integrating second layer outputs to the third layer;
• the third layer with a plurality of third layer nodes with a number of third layer nodes determined by at least one clustering algorithm on a set of values associated to each output attribute with one third layer node defined for at least one cluster, and links established between at least one third layer node and at least one node in the second layer for applying weights to the second layer outputs and passing them as third layer outputs to a fourth layer;
• the fourth layer with at least one fourth layer node with a number of fourth layer nodes determined by a number of output attributes, and links established between each third layer node and one fourth layer node for aggregating inputs from the fourth layer and for defuzzifying the third layer outputs to produce crisp output values;
• training of a neural network of the neuro-fuzzy expert system with the values associated to each input attribute and values associated to each output attribute so that a weight of at least one link between the second layer and third layer is adjusted based on a signal that comes from at least one third layer node and a signal from at least one second layer node.

A computer-implemented method for automatic generation of a neuro-fuzzy expert system, the computer-implemented method comprising:

• generating a neuro-fuzzy expert system from input data with a neural network structure comprising a plurality of layers including:
• a first layer with a plurality of first layer nodes with a number of first layer nodes determined by at least one clustering algorithm performed on a set of values associated to each input attribute with one first layer node defined for at least one cluster, and correspondence established between each first layer node and one input attribute for fuzzifying by determining each membership value of inputs, and passing each membership value as first layer outputs to a second layer;
• the second layer with a plurality of second layer nodes with a number of second layer nodes determined by at least one clustering algorithm on a set of values associated to each output attribute with one second layer node defined for at least one cluster, and correspondence established between each second layer node and one output attribute, and links established between at least one first layer node and at least one node in the second layer for applying weights to the first layer outputs and for aggregating first layer outputs to the second layer and for defuzzifying the first layer outputs to produce crisp output value;
• training of a neural network of the neuro-fuzzy expert system with the values associated to each input attribute and values associated to each output attribute so that a weight of at least one link between the first layer and second layer is adjusted based on a signal that comes from at least one first layer node and a signal from at least one second layer node;
• and
• presenting to a user of at least one result representing information embedded in a plurality of parameters and in at least one structure of the neuro-fuzzy expert system which has been trained.