Prosecution Insights
Last updated: November 04, 2026
Application No. 18/103,672

AUTOMATED SYNTHETIC AUGMENTATION OF MACHINE LEARNING MODEL TRAINING DATASETS AND INPUT VARIABLES TO IMPROVE ACCURACY

Final Rejection §101§103
Filed
Jan 31, 2023
Examiner
MULLINAX, CLINT LEE
Art Unit
2123
Tech Center
2100 — Computer Architecture & Software
Assignee
Evicore Healthcare Msi LLC
OA Round
2 (Final)
47%
Grant Probability
Moderate
3-4
OA Rounds
10m
Est. Remaining
84%
With Interview

Examiner Intelligence

Grants 47% of resolved cases
47%
Career Allowance Rate
63 granted / 133 resolved
-7.6% vs TC avg
Strong +37% interview lift
Without
With
+37.1%
Interview Lift
resolved cases with interview
Typical timeline
4y 7m
Avg Prosecution
17 currently pending
Career history
165
Total Applications
across all art units

Statute-Specific Performance

§101
21.7%
-18.3% vs TC avg
§103
54.5%
+14.5% vs TC avg
§102
7.3%
-32.7% vs TC avg
§112
13.3%
-26.7% vs TC avg
Black line = Tech Center average estimate • Based on career data from 133 resolved cases

Office Action

§101 §103
Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Status of Claims This action is responsive to the application filed on 08/07/2026. Claims 1-20 are pending. Claims 1, 4, 11, and 14 have been amended. Response to Arguments Applicant’s arguments, with respect to the rejection(s) of claim(s) 1-20 under 35 U.S.C. 101, have been considered but they are not persuasive. The applicant argues that the amended claims of training dataset manipulation and TF-IDF operations of converted training values, wherein the “claimed pipeline as a whole cannot practically be performed in the human mind” and “integrates any such idea into a practical application”. Therefore, the claims overcome the 101 rejections. The examiner respectfully disagrees. The additional elements and use in the claim do not operate to overcome the previous 101 abstract idea rejection. The training data manipulation limitations are deemed as performed in the human mind, and the TF-IDF operations are deemed as statistical calculations performed mentally or with the aid of pen and paper. Applicant is encouraged to give clarifying amendments to the machine learning model architecture/operations and/or the training process of the model to assist in overcoming the 101 rejections so the model is not merely a “black-box” form of computations. See 35 U.S.C 101 section for full, updated analysis of claim limitations necessitated by applicant amendments. Applicant’s arguments, with respect to the rejection(s) of claim(s) 1 and 11 under 35 U.S.C. 103, stating no reference teaches the amended limitations, since Farrugia “does not test training-data elements against the upper and lower bounds and assign those elements to bins based on the test”, “He does not perform TF-IDF filtering”, and motivation to combine Farrugia and He in view of the amendments have been considered but are moot because the arguments do not apply to the current combination of references being used in the current rejection. See 35 U.S.C 103 section for full mapping of claim limitations necessitated by applicant amendments. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claims 1 and 11 are respectively drawn to a system, method, and non-transitory computer readable storage medium, hence each falls under one of four categories of statutory subject matter (Step 1). Nonetheless, the claims are directed to a judicially recognized exception of an abstract idea without significantly more. Claims 1 and 11 recite the following, or analogous, limitations “loading a training data set, wherein: a first range is defined by an upper bound and a lower bound, the training data set includes a first bin and a second bin, loading includes assigning elements of the training data set having values within the first range to the first bin, and loading includes assigning elements of the training data set having values outside of the first range to the second bin, applying an under-sampling technique to elements of the first bin to generate an updated first bin, applying an over-sampling technique to elements of the second bin to generate an updated second bin, generating an updated training data set by merging the updated first bin and the updated second bin…; generating input variables by: assigning alphanumeric strings to elements of a raw data set, tokenizing the alphanumeric string, converting the tokenized strings to scalar values, performing term frequency-inverse document frequency filtering to generate filtered scalar values, wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects, and saving the filtered scalar values as input variables; and providing the input variables to the…model to generate output variables”. These limitations, as claimed, under its broadest reasonable interpretation, can be evaluated in a human mind, except for the recitation of generic computer components (using artificial intelligence/machine learning, a computer including one or more microprocessors, and a non-transitory computer readable storage medium) (Step 2A). Other than reciting “memory hardware”, “processing hardware”, “training a machine learning model by”, “and training the machine learning model with the updated training data set”, and “trained machine learning model” to perform the exceptions, nothing in the claims preclude the steps from practically being performed in the human mind. For example, a human expert can: mentally/with the aid of pen and paper loading a training data set, wherein: a first range is defined by an upper bound and a lower bound, the training data set includes a first bin and a second bin (e.g. by thinking of/writing out remembering a set of samples and dividing them into two categories based on majority and minority and upper and lower limits), mentally/with the aid of pen and paper loading includes assigning elements of the training data set having values within the first range to the first bin, and loading includes assigning elements of the training data set having values outside of the first range to the second bin (e.g. by thinking of/writing out a first category including the samples within the limits and a second category including samples outside the limits), mentally/with the aid of pen and paper applying an under-sampling technique to elements of the first bin to generate an updated first bin (e.g. by thinking of/writing out selecting a first subset of the majority category of samples by choosing fewer samples than the total), mentally/with the aid of pen and paper applying an over-sampling technique to elements of the second bin to generate an updated second bin (e.g. by thinking of/writing out selecting a second subset of the minority category of samples by choosing more that the first subset), mentally/with the aid of pen and paper generating an updated training data set by merging the updated first bin and the updated second bin… (e.g. by thinking of/writing out combining the first and second subset), mentally/with the aid of pen and paper generating input variables by: assigning alphanumeric strings to elements of a raw data set, tokenizing the alphanumeric string, converting the tokenized strings to scalar values (e.g. by thinking of/writing out associating numbers and letters labels to remembered samples, chunking the labels, and turning the chunks into scalar values), mentally/with the aid of pen and paper performing term frequency-inverse document frequency filtering to generate filtered scalar values, wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects, and saving the filtered scalar values as input variables (e.g. by thinking of/writing out a tfidf mathematical computation for weighting the scalars by determining a count of appearance for each category in a sample dataset, and remembering the results). mentally/with the aid of pen and paper providing the input variables to the…model to generate output variables (e.g. by thinking of/writing out a calculation to input the values for acquiring output data). Thus, the claims recite a mental process (Step 2A, Prong 1). Claims 1, 9, and 17 include additional elements, “memory hardware”, “processing hardware”, “training a machine learning model by”, “and training the machine learning model with the updated training data set”, and “trained machine learning model”, however the recitations of these elements are at a high level of generality, and amount to adding the words “apply it” (or an equivalent) with the judicial exception (i.e., “training a machine learning model by”, “and training the machine learning model with the updated training data set”), or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (i.e., “memory hardware”, “processing hardware”) (see MPEP 2106.05(f)); and generally linking the user of the judicial exception to a particular technological environment or field of use (i.e., “trained machine learning model”) (see MPEP 2106.05(h)). Hence, each of the additional limitations or in combination do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea (Step 2A, Prong 2; see MPEP 2106.05(f)). The additional elements in the claim do not amount to significantly more than an abstract idea. Furthermore, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional elements of using “memory hardware”, “processing hardware”, “training a machine learning model by”, “and training the machine learning model with the updated training data set”, and “trained machine learning model” to perform the steps of the independent claims amounts to no more than mere adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea; and generally linking the user of the judicial exception to a particular technological environment or field of use; as these cannot provide an inventive concept. (STEP 2B). As such, claims 1 and 11 are not patent eligible. Dependent claims 2-10, and 12-20 are also ineligible for the same reasons given with respect to claims 1 and 11. The dependent claims describe additional mental processes: mentally/with the aid of pen and paper automatically determining optimal hyperparameters for the…model; and configuring the…model with the optimal hyperparameters (claims 2 and 12) (e.g. by mentally/writing out determining and applying coefficients for the calculation) mentally/with the aid of pen and paper wherein determining the optimal hyperparameters includes: loading baseline hyperparameters; configuring the…model with the baseline hyperparameters; running the configured…model to determine baseline metrics; and in response to the baseline metrics being above a threshold, saving the baseline hyperparameters as the optimal hyperparameters (claims 3 and 13) (e.g. by mentally/writing out the calculation with initial coefficients, determining the output loss being greater than a predetermined value, and remembering the calculation coefficients) mentally/with the aid of pen and paper wherein determining the optimal hyperparameters includes, in response to the baseline metrics being at or below the threshold: adjusting the baseline hyperparameters, reconfiguring the…model with the adjusted hyperparameters, running the reconfigured…model to determine updated metrics, and in response to the updated metrics being more optimal than the baseline metrics, saving the adjusted hyperparameters as the optimal hyperparameters (claims 4 and 14) (e.g. by mentally/writing out determining the calculation output loss being less than a predetermined value, changing the calculation coefficients, determining the updated output loss being greater than a predetermined value, and remembering the changed calculation coefficients) mentally/with the aid of pen and paper in response to determining the output variables are above a threshold: loading a second…model, and providing the input variables to the second…model to generate second output variables (claims 5 and 15) (e.g. by thinking of/writing out determining the output loss being greater than a predetermined value, replacing the calculation with a second calculation, and determining a second output) mentally/with the aid of pen and paper in response to determining the output variables are not above the threshold: loading a third…model, and providing the input variables to the third…model to generate third output variables (claims 6 and 16) (e.g. by thinking of/writing out determining the output loss being less than a predetermined value, replacing the calculation with a third calculation, and determining a third output) mentally/with the aid of pen and paper wherein input variables include a third bin and a fourth bin (claims 9 and 19) (e.g. by thinking of/writing out the remembered samples having third and fourth categories) mentally/with the aid of pen and paper wherein the input variables of the third bin are assigned to fixed effects and the input variables of the fourth bin are assigned to mixed effects (claims 10 and 20) (e.g. by thinking of/writing out the third category and fourth category being fixed and mixed effects) Again, the dependent claims continued to cover the performance of the limitation in the mind as inherited from the independent claims (Step 2A, Prong 1). The dependent claims 2-6 and 12-16 recitation of “trained machine learning model” and “machine learning”; claims 7 and 17 recitation of “wherein the trained machine learning model includes a light gradient-boosting machine model”; claims 8 and 18 recitation of “wherein the trained machine learning model includes a mixed effects random forests model with a light gradient-boosting machine regressor” are again recited at a high level and amount to generally linking the user of the judicial exception to a particular technological environment or field of use, and do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea (Step 2A, Prong 2). The additional element in the claims do not amount to significantly more than an abstract idea. As discussed above with respect to the integration of the abstract idea into a practical application, the additional elements to perform the steps of in the dependent claims and perform the steps of the claims amount to no more than generally linking the user of the judicial exception to a particular technological environment or field of use, and this cannot provide an inventive concept. (STEP 2B). As such, dependent claims 2-10, and 12-20 additional elements or combination of elements do not amount to significantly more than an abstract idea nor provide any inventive concept, nor impose a meaningful limit to integrate the elements into a practical application or significantly more than the judicial exceptions; therefore, the dependent claims are not deemed patent eligible. Claim Rejections - 35 USC § 103 The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention. Claims 1-2, 7, 11-12, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Farrugia et al (“A Real‑Time Prescriptive Solution for Explainable Cyber‑Fraud Detection Within the iGaming Industry”, 2021) hereinafter Farrugia, Bhattacharyya et al (“Data mining for credit card fraud: A comparative study”, 2011) hereinafter Bhattacharyya, in view of Tal et al (US Pub 20230376858) hereinafter Tal, in view of He et al (“ADASYN: Adaptive Synthetic Sampling Approach for Imbalanced Learning”, 2008) hereinafter He, in view of Bykov et al (US Pub 20210065126) hereinafter Bykov. Regarding claims 1 and 11, Farrugia teaches a computer-implemented method comprising; and a system comprising: memory hardware configured to store instructions and processing hardware configured to execute the instructions (sections “Experimental Setup” teaches “We conducted our experiments using Jupyter Notebooks on a 16GB RAM, 64-bit Unix system” that includes one or more processors communicatively coupled to one or more memories executing code for performing the embodiments of the disclosure), wherein the instructions include: training a machine learning model by: loading a training data set, wherein: the training data set includes a first bin and a second bin, (section “Data Preparation”-“Data Pre-Processing” teach “Our dataset consisted of 451,123 players which included a total of 13,591 confirmed fraudsters, with the remaining players (437,532) not being previously flagged for fraud”), applying an under-sampling technique to elements of the first bin to generate an updated first bin (section “Data Preparation” teaches a labeled dataset of “non-fraud” and “fraud” samples, and “we filtered out those non-fraud players (first bin) who had no activity recorded, bringing down the total number of players to 197,733 (184,142 of which not previously flagged as fraud). Besides reducing noise, this process also acted as an under-sampling strategy and helped improve our class imbalance issue slightly” and bringing down the majority class.), applying an over-sampling technique to elements of the second bin to generate an updated second bin (sections “Data Preparation” and “Predictive Modeling” teach “We also experimented with over-sampling the minority class [fraud class] (second bin) using Synthetic Minority Over-sampling TEchnique (SMOTE) to balance both classes further”), generating an updated training data set by merging the updated first bin and the updated second bin (section “Data Preparation”-“Predictive Modelling” teach over and under sampling the different classes in the training dataset, then training machine learning “RF, LGB, DT, and LR [models] on our dataset”), and training the machine learning model with the updated training data set (section “Data Preparation”-“Predictive Modelling” teach over and under sampling the different classes in the training dataset, then training machine learning “RF, LGB, DT, and LR [models] on our dataset”); generating input variables by: assigning alphanumeric strings to elements of a raw data set (section “Data Processing” teaches “binary classification dataset (‘non-fraud’ and ‘fraud’)”), tokenizing the alphanumeric string (section “Data Preparation” teaches “We monitored over 1000 dimensions which we later reduced to 25 features based on the following attributes (we further discuss this process in Sect. 3.3): 1. Multi-session behavioural aggregates, 2. Gaming patterns, 3. Session identification and geolocation, 4. Demographics, 5. Payment information”; thus converting the labeled data into feature values), converting the tokenized strings to scalar values (section “Data Pre-processing” teaches “we categorised our features into three groups: Boolean’s, scalars, and highly-skewed scalar”), However, while Farrugia teaches two categorized training data samples, binning training data, categorizing features into scalar values by scaling the features “within its respective inter-quartile range”, and removing features based on a contribution threshold, Farrugia does not explicitly teach a first range is defined by an upper bound and a lower bound, the training data set includes a first bin and a second bin, loading includes assigning elements of the training data set having values within the first range to the first bin, and loading includes assigning elements of the training data set having values outside of the first range to the second bin, and performing term frequency-inverse document frequency filtering to generate filtered scalar values, wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects, and saving the filtered scalar values as input variables; and providing the input variables to the trained machine learning model to generate output variable. Bhattacharyya teaches a first range is defined by an upper bound and a lower bound, the training data set includes a first bin and a second bin, loading includes assigning elements of the training data set having values within the first range to the first bin, and loading includes assigning elements of the training data set having values outside of the first range to the second bin (section 5.1 teaches dividing a fraud training dataset “into two subsets of 1237 (51%) (first range is defined by an upper bound and a lower bound…training data set having values within the first range to the first bin) and 1183 (49%) (training data set having values outside of the first range to the second bin) transactions” of fraudulent transaction samples). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to implement Bhattacharyya’s teachings of training dataset subset division amounts into Farrugia’s teaching of over and under sampling of different data types for training models and input data conversion in order to create tailored training datasets for accurately training models (Bhattacharyya, sections 5.1-5.2). However, Bhattacharyya does not explicitly teach performing term frequency-inverse document frequency filtering to generate filtered scalar values, wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects. Tal teaches performing term frequency-inverse document frequency filtering to generate filtered scalar values (paragraphs 0079 teach “predictive data analysis computing entity 106 may utilize a plurality of different approaches (e.g., at least 3 different vectorization techniques) for translating raw text strings (e.g., a medical document) into vectors to determine which adds the most predictive value to the classification-based machine learning models that it trains. For example, the predictive data analysis computing entity 106 may process standardized, chunked text using language-based models, word embedding techniques, and term frequency inverse document frequency (TFIDF) representations”). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, to include TFIDF operations for determining text string vector occurrence as taught by Tal in order to improve efficiency and accuracy of model training (Tal, paragraphs 0015-0017 and 0078-0079). However, Tal does not explicitly teach wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects. He teaches performing…frequency filtering to generate filtered scalar values, wherein weights assigned to the filtered scalar values increase based on frequencies of corresponding scalar values within respective data objects and decrease based on frequencies of the corresponding scalar values across a group of data objects (sections 2-3 teach “Our objective here is similar to those in SMOTEBoost [16] and DataBoost-IM [17] algorithms: providing different weights (emphasize) for different minority examples (based on a frequency) to compensate for the skewed distribution” of the converted data including based on count of each example type), and saving the filtered scalar values as input variables (section 2 teaches weighting the training and testing data sets according to majority and minority of appearance); and providing the input variables to the trained machine learning model to generate output variable (sections 2 and 3C teach using the training and testing set on “a decision tree as the base classifier”). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, as modified by TFIDF operations for determining text string vector occurrence as taught by Tal, to include He’s teachings of weighting the training and testing data sets according to majority and minority of appearance in order to acquire “improved accuracy” for predictions on the minority and majority classes (He, sections 2 and 3C). Further, Farrugia at least implies tokenizing each alphanumeric string (see mappings above); however, Bykov teaches tokenizing each alphanumeric string (paragraphs 0092-0094 teach “clean the string (i.e., collapse multiple spaces), surround alphanumeric characters with spaces, tokenize the string, identify skill variants and wrap them in tags”). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, as modified by TFIDF operations for determining text string vector occurrence as taught by Tal, as modified by He’s teachings of weighting the training and testing data sets according to majority and minority of appearance, to include string data tokenization as taught by Bykov in order to more efficiently train the model from data conversion and segmentation (Bykov, paragraphs 0092-0099). Regarding claims 2 and 12, the combination of Farrugia, Bhattacharyya, Tal, He, and Bykov teach all the claim limitations of claims 1 and 11 above; and further teach automatically determining optimal hyperparameters for the trained machine learning model; and configuring the trained machine learning model with the optimal hyperparameters (Farrugia, section “Predictive Modelling” teach “We evaluated all models using their default hyper-parameters (from the scikit-learn Python package). We further performed hyper-parameter tuning using Bayesian optimization and distributed Tree of Parzen Estimators (TPE) of the best performing model”). Regarding claims 7 and 17, the combination of Farrugia, Bhattacharyya, Tal, He, and Bykov teach all the claim limitations of claims 1 and 11 above; and further teach wherein the trained machine learning model includes a light gradient-boosting machine model (Farrugia, sections “Data Pre-processing”-“Predictive Modelling” teach using a “LightGBM (LGB) model”). Claims 3-4 and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Farrugia et al (“A Real‑Time Prescriptive Solution for Explainable Cyber‑Fraud Detection Within the iGaming Industry”, 2021) hereinafter Farrugia, Bhattacharyya et al (“Data mining for credit card fraud: A comparative study”, 2011) hereinafter Bhattacharyya, in view of Tal et al (US Pub 20230376858) hereinafter Tal, in view of He et al (“ADASYN: Adaptive Synthetic Sampling Approach for Imbalanced Learning”, 2008) hereinafter He, in view of Bykov et al (US Pub 20210065126) hereinafter Bykov, in view of Mostafa et al (US Pub 20220284353) hereinafter Mostafa. Regarding claims 3 and 13, the combination of Farrugia, Bhattacharyya, Tal, He, and Bykov teach all the claim limitations of claims 2 and 12 above; and further teach wherein determining the optimal hyperparameters includes: loading baseline hyperparameters; configuring the trained machine learning model with the baseline hyperparameters; running the configured machine learning model to determine baseline metrics (Farrugia, section “Predictive Modelling” teach “We evaluated all models using their default hyper-parameters (from the scikit-learn Python package).); However, the combination does not explicitly teach and in response to the baseline metrics being above a threshold, saving the baseline hyperparameters as the optimal hyperparameters. Mostafa teaches and in response to the baseline metrics being above a threshold, saving the baseline hyperparameters as the optimal hyperparameters (paragraphs 0027-0029 and 0043 teach loading previously determined hyperparameters that satisfy a threshold “to configure a training algorithm which is then used to train on a local dataset” and determine if the model’s “loss has improved by an amount larger than a threshold (baseline metrics being above a threshold) compared to the loss of the aggregate model from the previous round”, and using the hyperparameters further in future rounds (saving the baseline hyperparameters as the optimal hyperparameters)). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, as modified by TFIDF operations for determining text string vector occurrence as taught by Tal, as modified by He’s teachings of weighting the training and testing data sets according to majority and minority of appearance, as modified by string data tokenization as taught by Bykov, to include hyperparameter performance comparison against a threshold for saving parameters for future use as taught by Mostafa in order to more efficiently use a computer system when training machine learning models (Mostafa, paragraphs 0027-0029, 0043, and 0110). Regarding claims 4 and 14, the combination of Farrugia, He, Bykov, and Mostafa teach all the claim limitations of claims 3 and 13 above; and further teach wherein determining the optimal hyperparameters includes, in response to the baseline metrics being at or below the threshold: adjusting the baseline hyperparameters, reconfiguring the trained machine learning model with the adjusted hyperparameters, running the reconfigured machine learning model to determine updated metrics, and in response to the updated metrics being more optimal than the baseline metrics, saving the adjusted hyperparameters as the optimal hyperparameters (Farrugia, section “Predictive Modelling” teach “We evaluated all models using their default hyper-parameters (from the scikit-learn Python package). We further performed hyper-parameter tuning using Bayesian optimization and distributed Tree of Parzen Estimators (TPE) of the best performing model”). Claims 5-6 and 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Farrugia et al (“A Real‑Time Prescriptive Solution for Explainable Cyber‑Fraud Detection Within the iGaming Industry”, 2021) hereinafter Farrugia, Bhattacharyya et al (“Data mining for credit card fraud: A comparative study”, 2011) hereinafter Bhattacharyya, in view of Tal et al (US Pub 20230376858) hereinafter Tal, in view of He et al (“ADASYN: Adaptive Synthetic Sampling Approach for Imbalanced Learning”, 2008) hereinafter He, in view of Bykov et al (US Pub 20210065126) hereinafter Bykov, in view of Alon et al (US Pub 20220156524) hereinafter Alon. Regarding claims 5 and 15, the combination of Farrugia, Bhattacharyya, Tal, He, and Bykov teach all the claim limitations of claims 1 and 11 above; however, the combination does not explicitly teach in response to determining the output variables are above a threshold: loading a second trained machine learning model, and providing the input variables to the second trained machine learning model to generate second output variables. Alon teaches in response to determining the output variables are above a threshold: loading a second trained machine learning model, and providing the input variables to the second trained machine learning model to generate second output variables (paragraph 0004 teaches determining a first model output and the output’s “correctness metric”, and “responsive to determining that the correctness metric exceeds the threshold: (a) applying the second machine learning model to the input to generate a second model output”). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, as modified by TFIDF operations for determining text string vector occurrence as taught by Tal, as modified by He’s teachings of weighting the training and testing data sets according to majority and minority of appearance, as modified by string data tokenization as taught by Bykov, to include model outputs compared to thresholds for model replacement as taught by Alon in order to obtain more accurate model predictions based on a performance metric (Alon, 0004 and 0027). Regarding claims 6 and 16, the combination of Farrugia, Bhattacharyya, Tal, He, Bykov, and Alon teach all the claim limitations of claims 5 and 15 above; and further teach in response to determining the output variables are not above the threshold: loading a third training machine learning model, and providing the input variables to the third trained machine learning model to generate third output variables (Alon, paragraph 0027 teaches “second correctness metric could be generated by second machine learning model 220b and/or determined from its output and the second correctness metric could be used to determine whether to conditionally execute further machine learning models (e.g., a third machine learning model). This could include comparing the second correctness metric to a second threshold”). Farrugia, Bhattacharyya, Tal, He, Bykov, and Alon are combinable for the same rationale as set forth above with respect to claims 5 and 15. Claims 8-10 and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Farrugia et al (“A Real‑Time Prescriptive Solution for Explainable Cyber‑Fraud Detection Within the iGaming Industry”, 2021) hereinafter Farrugia, Bhattacharyya et al (“Data mining for credit card fraud: A comparative study”, 2011) hereinafter Bhattacharyya, in view of Tal et al (US Pub 20230376858) hereinafter Tal, in view of He et al (“ADASYN: Adaptive Synthetic Sampling Approach for Imbalanced Learning”, 2008) hereinafter He, in view of Bykov et al (US Pub 20210065126) hereinafter Bykov, in view of Provencher et al (US Pub 20230077428) hereinafter Provencher. Regarding claims 8 and 18, the combination of Farrugia, Bhattacharyya, Tal, He, and Bykov teach all the claim limitations of claims 1 and 11 above; however, the combination does not explicitly teach wherein the trained machine learning model includes a mixed effects random forests model with a light gradient-boosting machine regressor. Provencher teaches wherein the trained machine learning model includes a mixed effects random forests model with a light gradient-boosting machine regressor (paragraphs 0062-0068 teach a “MERF model” utilizing regression gradient boosting trees). Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify over and under sampling of different data types for training models and input data conversion, as taught by Farrugia as modified by Bhattacharyya’s teachings of training dataset subset division amounts, as modified by TFIDF operations for determining text string vector occurrence as taught by Tal, as modified by He’s teachings of weighting the training and testing data sets according to majority and minority of appearance, as modified by string data tokenization as taught by Bykov, to include a specific model type for data processing as taught by Provencher in order to obtain more accurate predictions based on MERF model data advantageous “cluster data” processing performance (Provencher, 0004 and 0027). Regarding claims 9 and 19, the combination of Farrugia, Bhattacharyya, Tal, He, Bykov, and Provencher teach all the claim limitations of claims 8 and 18 above; and further teach wherein input variables include a third bin and a fourth bin (He, sections 2 and 3C teach using the training (third bin) and testing set (fourth bin) on “a decision tree as the base classifier”). Farrugia, Bhattacharyya, Tal, He, Bykov, and Provencher are combinable for the same rationale as set forth above with respect to claims 8 and 18. Regarding claims 10 and 20, the combination of Farrugia, Bhattacharyya, Tal, He, Bykov, and Provencher teach all the claim limitations of claims 9 and 19 above; and further teach wherein the input variables of the third bin are assigned to fixed effects and the input variables of the fourth bin are assigned to mixed effects (Provencher, paragraphs 0062-0068 teach a “MERF model” processing datasets of “fixed effect” and datasets of mixed random and fixed effect data). Farrugia, Bhattacharyya, Tal, He, Bykov, and Provencher are combinable for the same rationale as set forth above with respect to claims 8 and 18. Conclusion Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to CLINT MULLINAX whose telephone number is 571-272-3241. The examiner can normally be reached on Mon - Fri 8:00-4:30 PT. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey Shmatov can be reached on 571-270-3428. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /C.M./Examiner, Art Unit 2123 /ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Read full office action

Prosecution Timeline

Jan 31, 2023
Application Filed
May 07, 2026
Non-Final Rejection mailed — §101, §103
Aug 07, 2026
Response Filed
Oct 19, 2026
Final Rejection mailed — §101, §103 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12775887
BOOTSTRAP METHOD FOR CROSS-COMPANY MODEL GENERALIZATION ASSESSMENT
3y 9m to grant Granted Oct 27, 2026
Patent 12710927
SOFTWARE PROCESS MODIFICATION PLATFORM FOR COMPLIANCE
5y 5m to grant Granted Aug 18, 2026
Patent 12704934
SUGGESTING ACTIVITIES
4y 4m to grant Granted Aug 11, 2026
Patent 12664407
Resource-Efficient Attention in a Neural Network
5y 2m to grant Granted Jun 23, 2026
Patent 12665986
LIVE STYLE TRANSFER ON A MOBILE DEVICE
4y 7m to grant Granted Jun 23, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

3-4
Expected OA Rounds
47%
Grant Probability
84%
With Interview (+37.1%)
4y 7m (~10m remaining)
Median Time to Grant
Moderate
PTA Risk
Based on 133 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month