# US 12,602,610 B2 · Classification based on imbalanced dataset > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,602,610 B2 - Document ID: US12602610B2 - Status: Granted - Kind: B2 - Application number: 17/465,916 - Filed: 2021-09-03 - Published: 2023-03-09 - Granted: 2026-04-14 - Earliest claimed: 2021-09-03 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Jing Xu; Si Er Han; Xue Ying Zhang; Xiao Ming Ma; Ji Hui Yang - Web page: https://patentgrep.com/patent/US12602610B2 ## CPC - G06N 20/00: Machine learning - G06N 20/20: Ensemble learning - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Methods, systems, and computer program products perform classification based on an imbalanced dataset. In a method, machine learning models are generated based on positive samples included in an imbalanced dataset. An amount of the positive samples is less than an amount of negative samples that are included in the imbalanced dataset. Each sample in the positive and negative samples includes parameters. Multiple influential parameter groups are identified from the parameters for the positive samples, respectively. A final predictive model is determined based on the machine learning models and the multiple influential parameter groups. The final predictive model is used for classifying a sample as a positive type or a negative type. ## Claims ### Claim 1 A method comprising: generating, by one or more processors, multiple machine learning models based on multiple positive samples included in an imbalanced dataset, wherein an amount of the multiple positive samples is less than an amount of multiple negative samples that are included in the imbalanced dataset, and wherein each of the multiple positive samples and the multiple negative samples includes respective parameters; identifying, by the one or more processors, from the respective parameters, multiple influential parameter groups respectively for the multiple positive samples, wherein the multiple influential parameter groups respectively cause the multiple machine learning models to classify an associated one of the multiple positive samples as a positive type; determining, by the one or more processors, a model set that includes at least some of the multiple machine learning models; and inputting a first new sample into the model set such that the model set, in response, classifies the first new sample as a positive type or a negative type, wherein the model set compares the multiple influential parameter groups to target influential parameter groups as part of the classifying. ### Claim 2 (depends on 1) The method of claim 1, wherein the generating the multiple machine learning models comprises: obtaining, by the one or more processors, a training dataset based on a first positive sample of the multiple positive samples, wherein the training dataset includes at least one positive training sample and negative training samples; and generating, by the one or more processors, the machine learning models by training an initial machine learning model with the training dataset. ### Claim 3 (depends on 2) The method of claim 2, wherein the obtaining the training dataset comprises: obtaining, by the one or more processors, the negative training samples, wherein the negative training samples include a first negative sample from the multiple negative samples of the imbalanced dataset and include a second negative sample obtained via modifying, into a negative sample, a second positive sample from the multiple positive samples of the imbalanced dataset; and adding, by the one or more processors, the multiple negative training samples into the training dataset. ### Claim 4 (depends on 2) The method of claim 2, wherein the obtaining the training dataset comprises: obtaining, by the one or more processors, the at least one positive training sample based on the first positive sample; and assigning, by the one or more processors, a weight to the at least one positive training sample based on a ratio of an amount of the negative training samples and an amount of the at least one positive training sample. ### Claim 5 (depends on 1) The method of claim 1, further comprising: responsive to determining that a further positive sample is added into the imbalanced dataset, generating, by the one or more processors, a further machine learning model based on the further positive sample; identifying, by the one or more processors, a further influential parameter group for the further machine learning model from parameters of the further positive sample; and updating, by the one or more processors, the model set to include the further machine learning model and the further influential parameter group. ### Claim 6 (depends on 1) The method of claim 1, further comprising: responsive to determining that a target sample is received, determining, by the one or more processors, multiple types for the target sample as output from machine learning models included in the model set; and responsive to determining that all of the multiple types indicate the negative type, classifying, by the one or more processors, the target sample as the negative type. ### Claim 7 A method comprising: generating, by one or more processors, multiple machine learning models respectively based on multiple positive samples included in an imbalanced dataset, wherein an amount of the multiple positive samples is below an amount of multiple negative samples that are included in the imbalanced dataset, and wherein each of the multiple positive samples and the multiple negative samples includes respective parameters; inputting a first new sample into the multiple machine learning models to respectively produce predictions that classify the first new sample as a positive type or a negative type; responsive to determining that at least one of the predictions indicates the positive type, generating, by the one or more processors, a model set including at least one of the machine learning models whose prediction indicated the positive type; determining, by the one or more processors, an influential parameter group associated with a first machine learning model in the model set, wherein the influential parameter group causes the first machine learning model to classify an associated one of the multiple positive samples as the positive type; identifying, by the one or more processors, a target first influential parameter group for the first new sample with respect to the first machine learning model, wherein the target first influential parameter group causes the first machine learning model to classify the first new sample as the positive type; removing, by the one or more processors, the first machine learning model from the model set in response to a difference between the influential parameter group and the first influential parameter group exceeding a predefined threshold value, wherein the removing changes the model set to an updated model set and the updated model set contains a smaller number of machine learning models than the model set contains; and classifying, by the one or more processors, the first new sample based on the updated model set. ### Claim 8 (depends on 7) The method of claim 7, further comprising: determining, by the one or more processors, an influential parameter group associated with a second machine learning model in the model set, wherein the influential parameter group associated with the second machine learning model causes the second machine learning model to classify an associated one of the multiple positive samples as the positive type; identifying, by the one or more processors, a target second influential parameter group for the first new sample with respect to the second machine learning model, wherein the target second influential parameter group causes the second machine learning model to classify the second new sample as the positive type; and in response to a difference between the influential parameter group and the target second influential parameter group being less than a predefined threshold value, keeping the second machine learning model as part of the updated model set, wherein the second machine learning model is used in the classifying of the first new sample based on the updated model set. ### Claim 9 (depends on 7) The method of claim 7, wherein the classifying the target sample based on the updated model set comprises: responsive to determining that the updated model set is empty, classifying, by the one or more processors, the target sample as the negative type. ### Claim 10 (depends on 7) The method of claim 7, wherein the classifying the target sample based on the updated model set comprises: responsive to determining that the updated model set is not empty, classifying, by the one or more processors, the target sample as the positive type. ### Claim 11 A computer system comprising: a computer processor; and a computer-readable memory unit that stores instructions that cause the computer processor to perform operations comprising: generating multiple machine learning models respectively based on multiple positive samples included in an imbalanced dataset, wherein an amount of the multiple positive samples is below an amount of multiple negative samples that are included in the imbalanced dataset, and wherein each of the multiple positive samples and the multiple negative samples includes respective parameters; identifying from the respective parameters multiple influential parameter groups respectively for the multiple positive samples, wherein the multiple influential parameter groups respectively cause the multiple machine learning models to classify an associated one of the multiple positive samples as a positive type; and determining a model set that includes at least some of the machine learning models; and inputting a first new sample into the model set such that the model set, in response classifies the first new sample as a positive type or a negative type, wherein the model set compares the multiple influential parameter groups to target influential parameter groups as part of the classifying. ### Claim 12 (depends on 11) The system of claim 11, wherein the generating the multiple machine learning models comprises: obtaining a training dataset based on a first positive sample of the multiple positive samples, wherein the training dataset includes at least one positive training sample and negative training samples; and generating the multiple machine learning models by training an initial machine learning model with the training dataset. ### Claim 13 (depends on 12) The system of claim 12, wherein the obtaining the training dataset comprises: obtaining the negative training samples, wherein the negative training samples include a first negative sample from the multiple negative samples of the imbalanced dataset and a second negative sample obtained by modifying one of the multiple positive samples of the imbalanced dataset into a negative sample; and adding the negative training samples into the training dataset. ### Claim 14 (depends on 12) The system of claim 12, wherein the obtaining the training dataset comprises: obtaining the at least one positive training sample based on the first positive sample; and assigning a weight to the at least one positive training sample based on a ratio of an amount of the multiple negative training samples and an amount of the at least one positive training sample. ### Claim 15 (depends on 11) The system of claim 11, wherein the operations further comprise: responsive to determining that a target sample is received, determining multiple types for the target sample based on the machine learning models included in the model set; and responsive to determining that all of the multiple types indicate the negative type, classifying the target sample as the negative type. ### Claim 16 (depends on 11) The system of claim 11, wherein the operations further comprise: updating the model set to form an updated model set; and classifying a target sample based on the updated model set, wherein the classifying comprises responsive to determining that the updated model set is empty, classifying the target sample as the negative type. ### Claim 17 (depends on 11) The system of claim 11, wherein the operations further comprise: updating the model set to form an updated model set; and classifying a target sample based on the updated model set, wherein the classifying comprises responsive to determining that the updated model set is not empty, classifying the target sample as the positive type. ### Claim 18 A computer program product comprising: a computer readable storage medium; and program instructions stored in the computer readable storage medium, the program instructions executable by a processor to cause the processor to perform operations comprising: generating multiple machine learning models respectively based on multiple positive samples included in an imbalanced dataset, wherein an amount of the multiple positive samples is below an amount of multiple negative samples that are included in the imbalanced dataset, and wherein each of the multiple positive samples and the multiple negative samples includes respective parameters; identifying from the respective parameters multiple influential parameter groups respectively for the multiple positive samples, wherein the multiple influential parameter groups respectively cause the multiple machine learning models to classify an associated one of the multiple positive samples as a positive type; determining a model set that includes at least some of the multiple machine learning models; and inputting a first new sample into the model set such that the model set, in response, classifies the first new sample as a positive type or a negative type, wherein the model set compares the multiple influential parameter groups to target influential parameter groups as part of the classifying. ### Claim 19 (depends on 18) The product of claim 18, wherein the operations further comprise: responsive to determining that a target sample is received, determining multiple types for the target sample based on the machine learning models included in the model set; and responsive to determining that all of the multiple types indicate the negative type, classifying the target sample as the negative type. ### Claim 20 (depends on 18) The product of claim 18, wherein the operations further comprise: responsive to determining that at least one predictive type of multiple predictive types indicates the positive type for a target sample, identifying a target influential parameter group for the target sample; updating the model set based on a difference between the influential parameter group associated with the machine learning model that indicated the positive type and the target influential parameter group to form an updated model set; and classifying the target sample based on the updated model set. ### Claim 21 (depends on 18) The product of claim 18, wherein the generating the multiple machine learning models comprises: obtaining a training dataset based on a first positive sample of the multiple positive samples, wherein the training dataset includes at least one positive training sample and negative training samples; and generating the multiple machine learning models by training an initial machine learning model with the training dataset. ### Claim 22 (depends on 21) The product of claim 21, wherein the obtaining the training dataset comprises: obtaining the negative training samples, wherein the negative training samples include a first negative sample from the multiple negative samples of the imbalanced dataset and a second negative sample obtained by modifying a first positive sample of the multiple positive samples of the imbalanced dataset into a negative sample; and adding the negative training samples into the training dataset. ### Claim 23 (depends on 21) The product of claim 21, wherein the obtaining the training dataset comprises: obtaining the at least one positive training sample based on the first positive sample; and assigning a weight to the at least one positive training sample based on a ratio of an amount of the negative training samples and an amount of the at least one positive training sample.