公开(公告)号：US20250068982A1

公开(公告)日：2025-02-27

申请号：US18236754

申请日：2023-08-22

Applicant: Panasonic Intellectual Property Management Co., Ltd.

Inventor： Yizhou HUANG ,  Chandra Suwandi WIJAYA ,  Debdeep PAUL ,  Koji MIURA

IPC: G06N20/20

Abstract: According to an embodiment, a method for determining feature importance in an ensemble model including a plurality of Machine Learning (ML) models is disclosed. The method includes receiving a dataset comprising input features and a forecast result. The method also includes generating a ranking-based feature list based on the input features. Further, the method includes generating a feature importance output based on the ranking-based features lists. Furthermore, the method includes determining a weightage value corresponding to each of the plurality of ML models based on an accuracy value associated with the corresponding machine learning model. The method also includes determining a weightage-based feature importance value corresponding to each input feature corresponding to the feature importance output based on the determined weightage value corresponding to each ML model responsible for the corresponding input feature in the feature importance output.

公开(公告)号：US20230032011A1

公开(公告)日：2023-02-02

申请号：US17389042

申请日：2021-07-29

Applicant: PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO., LTD.

Inventor： Koji MIURA ,  Yukinori SASAKI ,  Akira MINEGISHI ,  Yizhou HUANG ,  Debdeep PAUL ,  Yongning YIN ,  Khai JUN KEK

IPC: G06Q10/04 ,  G06N3/08 ,  G06K9/62

Abstract: A system for generating a forecast including a classifier module for receiving from a user, at least one feature and classifying the at least one feature into a plurality of priority groups based on a user preference. The system further includes an artificial intelligence (AI) forecast module in communication with the classifier module for processing the plurality of priority groups with at least one feature. The AI forecast module derive a learning from classification of the at least one feature into the plurality of priority groups; and generate the forecast based on the learning.

公开(公告)号：US20250086389A1

公开(公告)日：2025-03-13

申请号：US18367310

申请日：2023-09-12

Applicant: Panasonic Intellectual Property Management Co., Ltd.

Inventor： Gayathri SARANATHAN ,  Nway Nway AUNG ,  Ariel BECK ,  Chandra Suwandi WIJAYA ,  Jianyu CHEN ,  Debdeep PAUL ,  Sahim YAMAURA ,  Koji MIURA

IPC: G06F40/279 ,  G06N20/00

Abstract: According to an embodiment, a method for generating textual features corresponding to text documents from a raw dataset is disclosed. The method includes preprocessing the text documents and determining topic probability scores (TPS) and confidence scores (CS) using unsupervised and supervised machine learning models, respectively. The combination of TPS and CS is used to generate a compound distribution score (CDS), which forms a comprehensive representation of the output of the machine learning models. The determined TPS, CS, and CDS are then used to generate a set of textual features, which serve as independent variables for a forecasting model.

公开(公告)号：US20240320694A1

公开(公告)日：2024-09-26

申请号：US18124302

申请日：2023-03-21

Applicant: Panasonic Intellectual Property Management Co., Ltd.

Inventor： Debdeep PAUL ,  Sahim YAMAURA ,  Chandra Suwandi WIJAUA

IPC: G06Q30/0202 ,  G06N7/01

CPC classification number: G06Q30/0202 ,  G06N7/01

Abstract: According to an embodiment, a method for generating and explaining trend forecast of a timeseries with measures of quality of explainability is disclosed. The method comprises receiving a target variable and a set of relevant feature(s) corresponding to the variable. The method comprises performing a classification for the target variable, wherein the classification indicates classifying the target variable into a one or more states. Further, the method comprises determining a state transition matrices for each timestamp and design appropriate functions to model and quantify the trend forecast via a state transition score. The state transition score indicates transition between the corresponding states, wherein states may be obtained through suitable encoding of the target variable, and generating and explaining trend forecast based on the state transition.

公开(公告)号：US20240119470A1

公开(公告)日：2024-04-11

申请号：US17955053

申请日：2022-09-28

Applicant: Panasonic Intellectual Property Management Co., Ltd.

Inventor： Debdeep PAUL ,  Chandra Suwandi Wijaya ,  Yizhou Huang ,  Khai Jun Kek ,  Koji Miura

IPC: G06Q30/02

CPC classification number: G06Q30/0202

Abstract: According to an embodiment, a method for generating a forecast of a timeseries is disclosed. The method comprises receiving a set of features comprising data and timeseries to be used by each of a plurality of prediction models for generating the forecast. Further, the method comprises generating using the set of features, a plurality of forecast results based on an ensemble of the plurality of prediction models. Furthermore, the method comprises optimizing the plurality of forecast results associated with a respective forecast module. Additionally, the method comprises probabilistically combining the outputs of the plurality of optimization modules. Moreover, the method comprises outputting a final forecast based on the combination of the at least two forecast results.