Granted Patent B1

US 12,724,982 · App. 18/963,917 · Granted Sep 1, 2026

Inventor: Haribaskar Subramaniam (Dublin, IE)

Assignee: ICON Clinical Research Limited

G06F40/40G06F40/205G06F40/284G06F40/30G06N3/045G06N3/0985G06N20/20G16H10/20

Abstract

Claims (61)

1 . A system comprising:

a first machine learning model that comprises an ensemble model of a plurality of machine learning classifiers, wherein each the plurality of machine learning classifiers is independently trained to make one or more predictions and a final output is determined based on a hard voting algorithm applied to the one or more predictions of the plurality of machine learning classifiers, wherein the ensemble model is an integrated classifier model combining the plurality of machine learning classifiers, having diverse inductive biases and diverse learning mechanisms, wherein each of the plurality of machine learning classifiers is hyperparameter tuned to optimize performance of the ensemble model of the plurality of machine learning classifiers;

a processor storing instructions in a non-transitory memory that, when executed, cause the processor to:

acquire input data from one or more data sources;

pre-process the input data by

removing noise and data inconsistencies from the input data to generate cleaned data;

splitting the cleaned data into one or more tokens;

assigning at least one Part-of-Speech (POS) tag to the one or more tokens to generate one or more tagged tokens;

applying lemmatization and performing Unicode normalization to the one or more tagged tokens to generate text data;

performing vectorization to convert the text data into feature vectors; and

performing feature extraction and feature selection on the feature vectors to select relevant features of a first training dataset and a first validation dataset, wherein the first training dataset and the first validation dataset are distinct and different, wherein the independent training of each of the plurality of machine learning classifiers of the first machine learning model is performed using the relevant features of the first training dataset; and

validate the first machine learning model using the relevant features of the first validation dataset.

2 . The system of claim 1 , wherein the input data comprises at least one of structured data, semi-structured data, and unstructured data.

3 . The system of claim 1 , wherein the system further comprises a second machine learning model.

4 . The system of claim 3 , wherein the processor is configured to pre-process actual data;

feed the pre-processed actual data to the ensemble model of the plurality of machine learning classifiers to classify one or more data points of the pre-processed actual data as one of a relevant data point and an irrelevant data point based on a predefined threshold value;

retrieve, using the first machine learning model, raw input data that corresponds to the one or more data points that are classified as the relevant data point, wherein the raw input data is the input data acquired from the one or more data sources; and

feed retrieved raw input data of outcomes from the first machine learning model as an input to the second machine learning model.

5 . The system of claim 4 , wherein the input data is pre-processed by performing one or more preprocessing techniques using Term Frequency-Inverse Document Frequency (TF-IDF).

6 . The system of claim 4 , wherein the predefined threshold value is initially set based on at least one of outcomes of historical input data and predicted by the first machine learning model.

7 . The system of claim 4 , wherein the processor is further configured to:

train the second machine learning model using a second training dataset; and

validate the second machine learning model using a second validation dataset.

8 . The system of claim 7 , wherein the second machine learning model is a self-learning model comprising a feedback layer that enables the second machine learning model to learn continuously from the retrieved raw input data.

9 . The system of claim 7 , wherein the processor is configured to extract, using the second machine learning model, one or more clinical outcome assessment names from the retrieved raw input data that are classified as relevant based on training provided to the second machine learning model.

10 . The system of claim 7 , wherein the processor is further configured to generate an output comprising harmonized clinical outcome assessment names.

11 . The system of claim 3 , wherein the second machine learning model is a Large Language model (LLM).

12 . The system of claim 1 , wherein the system is further configured to:

customize the ensemble model of the plurality of machine learning classifiers by performing

manipulating the pre-processed input data to generate modified input data,

adding, modifying, or removing at least one node of the ensemble model of the plurality of machine learning classifiers, and

training the ensemble model of the plurality of machine learning classifiers using the modified input data.

13 . The system of claim 1 , wherein the plurality of machine learning classifiers comprises two or more of a Random Forest classifier model, an XGBoost model, a Support Vector Machine (SVM) model, a Stochastic Gradient Descent model, and a Logistic Regression model.

14 . The system of claim 1 , wherein the hard voting algorithm analyzes the one or more predictions from the plurality of machine learning classifiers and assigns a final class label based on a class predicted by a majority of the classifiers.

15 . The system of claim 1 , wherein the system is configured for tracking of a clinical outcome assessment.

16 . The system of claim 1 , wherein the first machine learning model is calibrated, selected, and ensembled from the plurality of machine learning classifiers evaluated against a set of performance criteria for tracking a clinical outcome assessment.

17 . The system of claim 1 , wherein the first machine learning model has a feedback loop, wherein an output from a previous step is fed back to the model in real-time to improve performance and accuracy of the output of a next step.

18 . A method comprising:

acquiring input data from one or more data sources;

pre-processing the input data by

removing noise and data inconsistencies from the input data to generate cleaned data;

splitting the cleaned data into one or more tokens;

assigning at least one Part-of-Speech (POS) tag to the one or more tokens to generate one or more tagged tokens;

applying lemmatization and performing Unicode normalization to the one or more tagged tokens to generate text data;

performing vectorization to convert the text data into feature vectors; and

performing feature extraction and feature selection on the feature vectors to select relevant features of a first training dataset and a first validation dataset,

wherein the first training dataset and the first validation dataset are distinct and different;

training independently each machine learning classifier of a plurality of machine learning classifiers of a first machine learning model, wherein the first machine learning model comprises an ensemble model of the plurality of machine learning classifiers, wherein each of the plurality of machine learning classifiers is independently trained to make one or more predictions and a final output is determined based on a hard voting algorithm applied to the one or more predictions of the plurality of machine learning classifiers, wherein the ensemble model is an integrated classifier model combining the plurality of machine learning classifiers, having diverse inductive biases and diverse learning mechanisms, wherein the independent training of each of the plurality of machine learning classifiers of the first machine learning model is performed using the relevant features of the first training dataset, wherein each of the plurality of machine learning classifiers is hyperparameter tuned to optimize performance of the ensemble model of the plurality of machine learning classifiers; and

validating the first machine learning model using the relevant features of the first validation dataset.

19 . A non-transitory computer-readable medium having stored thereon instructions executable by a computer system to perform operations comprising:

acquiring input data from one or more data sources;

pre-processing the input data by

removing noise and data inconsistencies from the input data to generate cleaned data;

splitting the cleaned data into one or more tokens;

assigning at least one Part-of-Speech (POS) tag to the one or more tokens to generate one or more tagged tokens;

applying lemmatization and performing Unicode normalization to the one or more tagged tokens to generate text data;

performing vectorization to convert the text data into feature vectors; and

performing feature extraction and feature selection on the feature vectors to select relevant features of a first training dataset and a first validation dataset,

wherein the first training dataset and the first validation dataset are distinct and different;

training independently each machine learning classifier of a plurality of machine learning classifiers of a first machine learning model, wherein the first machine learning model comprises an ensemble model of the plurality of machine learning classifiers, wherein each of the plurality of machine learning classifiers is independently trained to make one or more predictions and a final output is determined based on a hard voting algorithm applied to the one or more predictions of the plurality of machine learning classifiers, wherein the ensemble model is an integrated classifier model combining the plurality of machine learning classifiers, having diverse inductive biases and diverse learning mechanisms, wherein the independent training of each of the plurality of machine learning classifiers is performed using the relevant features of the first training dataset, wherein each of the plurality of machine learning classifiers is hyperparameter tuned to optimize performance of the ensemble model of the plurality of machine learning classifiers; and

validating the first machine learning model using the relevant features of the first validation dataset.

Assignments (1)

References Cited (48)