Method for predicting business performance using machine learning and apparatus using the same

A company business performance predicting apparatus comprising:

• a communication unit;
• a memory;
• and
• a processor,
• wherein the processor is configured to:
• obtain a data set including internal data and external data of a specific company from the memory through an information collection unit;
• classify the data set into first data including structured data and second data including unstructured data through an information classification unit;
• process the second data into third data being structured data corresponding to the first data through an information processing unit;
• extract a feature based on the first data and the third data through a feature extraction unit by performing converting raw data into the feature through a data mining technique;
• and
• provide, to an external device, business performance prediction data of the specific company and business indicator data of the specific company mapped to a predetermined statistical attribute field through a predetermined prediction model in response to an individual characteristic of the feature,
• wherein the processor is further configured to, in extracting of the feature:
• perform feature vectorization on the first data and the third data in consideration of a word extraction frequency in the first data and the third data and a weight according to the frequency;
• and
• after the feature vectorization, perform feature scaling through re-scaling and standard regularization on the first data and the third data,
• wherein the processor is further configured to, in performing of the feature vectorization, extract the feature from a processed text, and assign a vector value to the extracted feature,
• wherein, in extracting of the features, when the feature vectorization is processed through a bag-of-word (BOW) model, a data set in a form of a sparse matrix is generated, and a vector consisting of an occurrence frequency value is generated by assigning a high weight to a word having a high occurrence frequent in the data set, wherein the weight is set in proportion to the occurrence frequency and is determined by a count value, and
• wherein the processor is further configured to, in providing of the business performance prediction data:
• provide the business performance prediction data through an ensemble model, by using Equation (1) below:
• y
• ^
• 1
• =
• ∑
• k
• =
• 1
• k
• f
• k
• (
• x
• i
• )
• ,
• f
• k
• ∈
• F
• [
• Equation
• ⁢
• 1
• ]
• where ŷl denotes a predicted value, which is the business performance prediction data, Xi denotes an input value, which is the extracted feature, and fk denotes a classification and regression trees (CART) model.

A non-transitory computer-readable recording medium storing a program for performing following operations for performing a method performed by a company business performance predicting apparatus, when executed by at least one processor, wherein the operations includes:

• obtaining, by an information collection unit, a data set including internal data and external data of a specific company;
• classifying, by an information classification unit, the data set into first data including structured data and second data including unstructured data;
• processing, by an information processing unit, the second data into third data being structured data corresponding to the first data;
• extracting, by a feature extraction unit, a feature based on the first data and the third data by performing converting raw data into the feature through a data mining technique;
• and
• providing, to an external device, business indicator data of the specific company by mapping to business performance prediction data of the specific company and a predetermined statistical attribute field through a predetermined prediction model in response to an individual characteristic of the feature,
• wherein the extracting of the feature includes:
• performing feature vectorization on the first data and the third data in consideration of a word extraction frequency in the first data and the third data and a weight according to the frequency;
• and
• after the feature vectorization, performing feature scaling through re-scaling and standard regularization on the first data and the third data,
• wherein the performing of the feature vectorization comprises extracting the feature from a processed text, and assigning a vector value to the extracted feature,
• wherein the extracting of the features comprises:
• when the feature vectorization is processed through a bag-of-word (BOW) model, a data set in a form of a sparse matrix is generated, and a vector consisting of an occurrence frequency value is generated by assigning a high weight to a word having a high occurrence frequent in the data set, wherein the weight is set in proportion to the occurrence frequency and is determined by a count value, and
• wherein the providing of the business performance prediction data comprises:
• providing the business performance prediction data through an ensemble model, by using Equation (1) below:
• y
• ^
• 1
• =
• ∑
• k
• =
• 1
• k
• f
• k
• (
• x
• i
• )
• ,
• f
• k
• ∈
• F
• [
• Equation
• ⁢
• 1
• ]
• where ŷl denotes a predicted value, which is the business performance prediction data, Xi denotes an input value, which is the extracted feature, and fk denotes a classification and regression trees (CART) model.