Tuning of machine learning models using accuracy metrics selected to increase performance

A method, using at least one processor and memory configured with executable instructions that:

• under control of at least one processor and memory configured with executable instructions that:
• receive a plurality of predictive weights associated with a plurality of metrics to measure predictive accuracy that include false positive outcomes, false negative outcomes, true positive outcomes, and true negative outcomes;
• select a weight adjusted classification threshold indicating an amount of error tolerated in a computing application, for use in a classification model of a machine learning model in a service provider environment hosting the computing application, according to the plurality of predictive weights to enable the machine learning model to increase a total value of the machine learning model and decrease performance outcome errors;
• receive adjusted values for the plurality of predictive weights, the adjusted values for the plurality of predictive weights indicating a modified value for one of false positive predictive outcomes or false negative predictive outcomes;
• adjust the weight adjusted classification threshold based at least in part on the adjusted values for the plurality of predictive weights, including one of increasing the weight adjusted classification threshold when the adjusted values for the predictive weights indicate an increased weight for false positive predictive outcomes or a decreased weight for false negative predictive outcomes, or decreasing the weight adjusted classification threshold when the adjusted values for the predictive weights indicate an increased weight for false negative predictive outcomes or a decreased weight for false positive predictive outcomes;
• and
• classify a data point according to the weight adjusted classification threshold in the classification model.

At least one machine readable storage medium having instructions embodied thereon, the instructions when executed cause one or more processors and memory to:

• receive a plurality of predictive weights associated with a plurality of metrics to measure predictive accuracy that include false positive outcomes, false negative outcomes, true positive outcomes, and true negative outcomes;
• represent each of the plurality of predictive weights as acceptable performance outcome errors in a classification model;
• select a weight adjusted classification threshold indicating an amount of error tolerated in a computing application, for use in the classification model of a machine learning model in a service provider environment hosting the computing application, according to the plurality of predictive weights to enable the machine learning model to increase a total value of the machine learning model and decrease performance outcome errors;
• receive adjusted values for the plurality of predictive weights, the adjusted values for the plurality of predictive weights indicating a modified value for one of false positive predictive outcomes or false negative predictive outcomes;
• adjust the weight adjusted classification threshold based at least in part on the adjusted values for the plurality of predictive weights, including one of increasing the weight adjusted classification threshold when the adjusted values for the predictive weights indicate an increased weight for false positive predictive outcomes or a decreased weight for false negative predictive outcomes, or decreasing the weight adjusted classification threshold when the adjusted values for the predictive weights indicate an increased weight for false negative predictive outcomes or a decreased weight for false positive predictive outcomes;
• and
• classify a data point according to the weight adjusted classification threshold in the classification model.