A method of a computer system, wherein the computer system comprises a first computer, a second computer and a third computer connected via a network, the method comprising:
- storing, by the first computer, a prediction model which implements artificial intelligence (AI) for predicting an event of an object based on evaluation data that includes a plurality of feature values of respective features indicating a state of the object;
- storing, by the second:
- a database that stores a plurality of pieces of predetermined analysis data including a plurality of features identical to the evaluation data and corresponding feature values, and
- a feature profiling database that defines a change rule of each of the plurality of feature values included in the evaluation data, which is generated based on a result of the prediction of the event of the object by the AI prediction implemented by the prediction model, the change rule including an influence value range indicating a range of values obtainable by the features, and a variation unit indicating a minimum value a respective feature value varies, for each of the features;
- executing, by the first computer:
- calculating a contribution vector for each piece of the predetermined analysis data based on the predetermined analysis data and the prediction model, each contribution vector including a numerical degree of influence of each feature value for each piece of the predetermined analysis data;
- and
- sending the calculated contribution vectors to the second computer;
- executing, by the second computer:
- receiving the calculated contribution vectors from the second computer;
- calculating a range defined by a maximum value and a minimum value of the features values of each feature of the analysis data as a value range;
- calculating a minimum value of respective differences between the feature values of each feature as a variation unit;
- calculating an influence value range of each feature, the influence value range includes feature values of the predetermined analysis data having a respective numerical degree of influence from the corresponding contribution vector that is greater than a threshold value;
- and
- updating the feature profiling database by updating the influence value range with the calculated influence value range of each feature and updating the calculated variation unit of each feature,
- executing, by the third computer:
- generating a plurality of pieces of change policy data, which are changes of the plurality of feature values included in the evaluation data based on the feature profiling database;
- calculating an evaluation value of each of the plurality of pieces of change policy data, which indicates a magnitude of an effect each of the plurality of pieces of change policy data has on a key performance indicator (KPI) using the prediction model, the evaluation data, and the plurality of pieces of change policy data;
- generating display data indicating the plurality of pieces of change policy data and the evaluation value of each of the plurality of pieces of change policy data;
- and
- displaying the display data.