Observation target selection device, observation target selection method, and program
The observation target selection device uses machine learning to select and predict labels for data with missing values, improving accuracy and reducing observation costs by targeting specific observations.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- NIPPON TELEGRAPH & TELEPHONE CORP
- Filing Date
- 2025-04-10
- Publication Date
- 2026-07-29
AI Technical Summary
Conventional methods fail to select observation targets indicated by missing values according to individual data, leading to reduced accuracy in label prediction and increased costs due to unnecessary observations.
An observation target selection device using machine learning-trained models to select and predict labels for data with missing values, allowing for targeted observations based on individual data characteristics.
Enables accurate label prediction and reduces the number of necessary observations, thereby minimizing costs and patient burden.
Smart Images
Figure 0007896724000014 
Figure 0007896724000015 
Figure 0007896724000016
Abstract
Description
[Technical Field]
[0001] This disclosure pertains to an observation target selection device, an observation target selection method, and a program. [Background technology]
[0002] Generally, in the field of machine learning, if data to be labeled contains many missing values, the accuracy of predicting the labels for that data will be low. To improve accuracy, it is necessary to observe more subjects and reduce the number of missing values, but observation can be costly. For example, in medical testing, testing many test items (subjects of observation) is costly and also increases the burden on the patient. In such cases, a selection method for choosing a predetermined subject of observation from untested items (multiple candidate subjects of observation indicated by missing values) has already been proposed (Non-patent reference 1). [Prior art documents] [Non-patent literature]
[0003] [Non-Patent Document 1] Guyon, Isabelle, and Andr Elisseeff. "An introduction to variable and feature selection." Journal of Machine Learning Research 3.Mar (2003): 1157-1182. [Overview of the project] [Problems that the invention aims to solve]
[0004] However, conventional methods have the problem that they cannot select observation targets indicated by missing values according to individual data (for example, data on individual patients) because they select pre-fixed observation targets from among multiple observation targets indicated by missing values.
[0005] This invention has been made in view of the above points, and aims to select observation targets indicated by missing values according to individual data. [Means for solving the problem]
[0006] To solve the above problems, the invention according to claim 1 is an observation target selection device for selecting an observation target to be observed in the prediction phase, comprising: an input unit for inputting data including predetermined missing values due to unobserved data; a machine learning-trained selection model including data showing the characteristics of observed observation targets, which is used to select a predetermined observation target from among a plurality of observation target candidates indicated by the predetermined missing values based on the data including the predetermined missing values due to unobserved data; a prediction unit for selecting a specific observation target based on data including a specific missing value due to unobserved data; and an output unit for outputting selection result data showing the specific observation target selected by the prediction unit, wherein the prediction unit uses a machine learning-trained prediction model for predicting the labels of the data including the predetermined missing values. Based on the data including the predetermined missing values input by the input unit, a label is predicted. The output unit outputs prediction result data showing the label predicted by the prediction unit. The prediction unit predicts the label again based on the updated data, which is obtained by reflecting the characteristics of the specific observation target obtained from new observations based on the selected data into the predetermined missing values, and the output unit outputs prediction result data showing the label predicted again by the prediction unit. This is a device for selecting the object to be observed. [Effects of the Invention]
[0007] As described above, the present invention has the effect of allowing the selection of observation targets indicated by missing values according to the individual data. [Brief explanation of the drawing]
[0008] [Figure 1] This is a schematic diagram of the communication system of this embodiment. [Figure 2] This is a hardware configuration diagram of the observation target selection device and communication terminal. [Figure 3] This is a functional configuration diagram of an observation target selection device according to an embodiment of the present invention. [Figure 4] This flowchart shows the processing or operation during the learning phase of the observation target selection device according to this embodiment. [Figure 5] It is a flowchart showing the processing or operation in the prediction phase of the observation target selection device according to the present embodiment. [Figure 6] It is a diagram showing the evaluation result.
Embodiments for Carrying Out the Invention
[0009] Hereinafter, embodiments of the present invention will be described based on the drawings.
[0010] 〔System Configuration of Embodiment〕 First, the outline of the configuration of the communication system 1 of the present embodiment will be described using FIG. 1. FIG. 1 is a schematic diagram of a communication system according to an embodiment of the present invention.
[0011] As shown in FIG. 1, the communication system 1 of the present embodiment is constructed by an observation target selection device 3 and a communication terminal 5. The communication terminal 5 is managed and used by the user Y.
[0012] In addition, the observation target selection device 3 and the communication terminal 5 can communicate via a communication network 100 such as the Internet. The connection form of the communication network 100 may be either wireless or wired.
[0013] The observation target selection device 3 is composed of one or more computers. When the observation target selection device 3 is composed of a plurality of computers, it may be referred to as an "observation target selection device" or a "label prediction system".
[0014] The observation target selection device 3 is a computer, and by machine learning, even if the input data is data including missing values, it selects a new observation target according to each individual data and predicts the label of the data including missing values. Therefore, the observation target selection device 3 can also be referred to as a label prediction device.
[0015] For example, when the observation target selection device 3 is used for medical examinations, the observation target selection device 3 uses a machine learning-trained prediction model g2 to predict the label (e.g., the patient's disease name) of data including missing values due to unexamined items. Further, the observation target selection device 3 uses a machine learning-trained selection model f2 including data indicating the already observed features (e.g., the "value" of height) of the patient to select an unexamined item (e.g., weight) that is a specific observation target among a plurality of observation target candidates (unexamined items) indicated by the missing values. Thereby, the doctor performs the next examination of the patient referring to this selected unexamined item. Then, the observation target selection device 3 predicts again the label (such as the patient's disease name) of the data including the remaining missing values (e.g., blood pressure, blood, etc.) including the features (e.g., the "value" of weight) of the examined item next. By repeating such processing and operations, the observation target selection device 3 can select examination items according to each patient and predict the disease name of the patient, so that finally the disease name of each patient can be determined with as few examinations as possible.
[0016] The communication terminal 5 is a computer. In FIG. 1, as an example, a notebook personal computer is shown, but it is not limited to the notebook type and may be a desktop personal computer. Further, the communication terminal may be a smartphone or a tablet terminal. In FIG. 1, the user Y is operating the communication terminal 5.
[0017] 〔Hardware Configuration of Observation Target Selection Device and Communication Terminal〕 Next, the hardware configurations of the observation target selection device 3 and the communication terminal 5 will be described using FIG. 2. FIG. 2 is a hardware configuration diagram of the observation target selection device and the communication terminal.
[0018] As shown in FIG. 2, the observation target selection device 3 includes a processor 301, a memory 302, an auxiliary storage device 303, a connection device 304, a communication device 305, and a drive device 306. Each hardware component constituting the observation target selection device 3 is interconnected via a bus 307. <0,
[0019] The processor 301 acts as a control unit that controls the entire observation target selection device 3 and has various computing devices such as a CPU (Central Processing Unit). The processor 301 reads various programs into memory 302 and executes them. The processor 301 may also include GPGPU (General-purpose computing on graphics processing units).
[0020] Memory 302 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 301 and memory 302 form a so-called computer, and the computer realizes various functions by having the processor 301 execute various programs read from memory 302.
[0021] The auxiliary storage device 303 stores various programs and various information used when those programs are executed by the processor 301.
[0022] The connection device 304 is a connection device that connects an external device (for example, a display device 310, an operating device 311) to the observation target selection device 3.
[0023] The communication device 305 is a communication device for sending and receiving various types of information with other devices.
[0024] The drive device 306 is a device for setting the recording medium 330. The recording medium 330 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs (Compact Disc Read-Only Memory), flexible disks, and magneto-optical disks. The recording medium 330 may also include semiconductor memories that record information electrically, such as ROMs (Read Only Memory) and flash memory.
[0025] The various programs to be installed on the auxiliary storage device 303 are installed, for example, when the distributed recording medium 330 is set in the drive device 306 and the various programs recorded on the recording medium 330 are read by the drive device 306. Alternatively, the various programs to be installed on the auxiliary storage device 303 may be installed by downloading them from the network via the communication device 305.
[0026] Furthermore, Figure 2 shows the hardware configuration of communication terminal 5, but since the configuration is the same except for the change in code from the 300s to the 500s, we will omit the explanation of these.
[0027] [Functional configuration of the observation target selection device] Next, the functional configuration of the observation target selection device will be explained using Figure 3. Figure 3 is a functional configuration diagram of the observation target selection device according to this embodiment.
[0028] In Figure 3, the observation target selection device 3 includes an input unit 31, a prediction model learning unit 32, a selection model learning unit 33, a prediction unit 34, and an output unit 39. Each of these units is a function implemented by instructions from the processor 301 in Figure 2 based on a program.
[0029] The input unit 31 receives input from user Y's communication terminal 5, including learning data D that shows the characteristics of observed objects (such as height) (such as test results indicating height of 170 cm), and data that includes missing values.
[0030] The prediction model learning unit 32 performs machine learning on the prediction model g1 based on the training data D input by the input unit 31.
[0031] The selection model learning unit 33 performs machine learning on the selection model f1 based on data that includes missing values.
[0032] The prediction unit 34 uses a trained selection model f2 to sequentially select observation targets from among the candidate observation targets showing missing values, and uses a trained prediction model g2 to predict the labels of the data containing missing values. In the case of medical tests, observation targets include untested items showing missing values (such as blood tests) and tested features (such as height and weight).
[0033] The output unit 39 outputs the results predicted by the prediction unit 34 as result data to an external device (such as a communication terminal 5 or display device 310) of the observation target selection device 3.
[0034] The prediction model learning unit 32, the selection model learning unit 33, and the prediction unit 34 will be explained in detail later.
[0035] [Processing or operation of the embodiment] Next, the processing or operation of this embodiment will be described in detail with reference to Figures 4 to 5.
[0036] <Learning Phase> First, Figure 4 will be used to explain the processing or operation of the observation target selection device 3 during the learning phase. Figure 4 is a flowchart showing the processing or operation of the observation target selection device during the learning phase according to this embodiment.
[0037] First, the input unit 31 receives learning data D, which represents the characteristics of observed objects, from the user Y's communication terminal 5, operating device 311, etc. (S11).
[0038] This training data D is
number
number
[0039] Next, the prediction model learning unit 32 learns the prediction model g1 based on the training data D. Predictive model g1 is a model for predicting labels on data that may contain (or may contain) missing values.
number
number
[0040] The prediction model g1 is trained by the prediction model learning unit 32 to fit the training data D. For example, in the case of a classification problem with discrete labels, the prediction model learning unit 32 can learn by minimizing the cross-entropy error between the predicted value and the true value. The prediction model learning unit 32 can improve its prediction performance when there are missing values by artificially changing the training data D to include missing values using a random mask vector and learning to reduce the expected error for the changed data. Alternatively, the prediction model learning unit 32 can change the data to include artificial missing values by having the selection model learning unit 33 sequentially observe the missing values using the selection model f1 described later, instead of randomly.
[0041] Next, the input unit 31 receives data, including missing values, from the user Y's communication terminal 5, operating device 311, etc. (S13).
[0042] Next, the selection model learning unit 33 learns the selection model f1 based on the data, including missing values. The selected model is f1.
number
number
[0043] Any model can be used as the selected model f1, such as a neural network or a linear model. If the prediction model learning unit 32 improves its prediction performance when observing the d-th observation target, but does not improve when observing the d'-th observation target, it will choose a score (S) such that the d-th observation is higher. nd > S nd' Train the selected model f1 to output ).
[0044] For example, such learning is possible by maximizing the objective function shown in (Equation 1) and (Equation 2) below.
number
number
number
number
[0045] When the costs of inspections and the like required for observation are different, it is also possible to use an objective function weighted by the cost. For example, when the cost of the d-th feature is cd, [Number] is used as the objective function. As a result, for example, the higher the inspection cost, the less likely the inspection item or the like is to be selected as the observation target.
[0046] Thus, the processing or operation in the learning phase ends.
[0047] [Prediction Phase] Subsequently, using FIG. 5, the processing or operation in the prediction phase of the observation target selection device 3 will be described. FIG. 5 is a flowchart showing the processing or operation in the prediction phase of the observation target selection device according to the present embodiment.
[0048] First, the input unit 31 inputs data including a specific missing value from the communication terminal 5, the operation device 311, etc. (S21).
[0049] Next, the prediction unit 34 uses a machine learning-trained prediction model g2 to predict the labels of data containing predetermined missing values, and predicts the labels of the data containing missing values (for example, the patient's disease name) based on the specific data containing missing values input by the input unit 31 (S22). Then, the output unit 39 outputs the prediction result data predicted by the prediction unit 34 to an external device (such as a communication terminal 5 or display device 310) outside the observation target selection device 3 (S23).
[0050] Furthermore, the prediction unit 34 uses a machine learning-based selection model f2, which includes data showing the characteristics of observed objects (e.g., the "value" of height), to select a specific object to be observed (e.g., weight) from among the specific missing values in the data including missing values input by the input unit 31 (S24). Then, the output unit 39 outputs the data of the selection result (e.g., weight) selected by the prediction unit 34 to an external device (communication terminal 5, display device 310, etc.) (S25). As a result, user Y (e.g., a doctor) can predict the label of the data including missing values (e.g., the patient's disease name) and understand which object to observe next (e.g., weight).
[0051] Next, the prediction unit 34 determines whether the termination conditions are met (S24). Termination conditions include the number of selections exceeding a certain value, the number of observation targets exceeding a certain value, the number of unobserved observation targets falling below a certain value, and the fluctuation of the predicted value falling below a certain value (for example, when the predicted disease name does not change even if predictions are repeated, making further predictions pointless). If the termination conditions are not met (S26: NO), user Y (for example, a doctor) observes a new specific observation target (for example, weight) based on the selection results (for example, weight) from the above process (S25) (for example, weight measurement), and the input unit 31 returns to process (S21) and inputs data that includes the characteristics of this specific observation target (for example, the "value" of weight) and the remaining missing values which are unobserved observation targets other than this specific observation target (for example, blood pressure, blood). As a result, in processing (S22), the prediction unit 34 uses the trained prediction model g2 to predict labels (for example, the patient's disease name), and in processing (S23), it outputs the prediction result data. In this case, since the number of observed subjects has increased, new prediction result data may be output.
[0052] Furthermore, in processing again (S24), the prediction unit 34 uses the trained selection model f2 to select an observation target (e.g., blood pressure) from the remaining missing values. Then, in processing again (S25), the output unit 39 outputs the selected data (e.g., blood pressure). In this way, the prediction unit 34 performs sequential processing by repeating processing (S22, S24), and can select an observation target according to the data of each individual patient (e.g., disease name). In addition, this makes it possible to grasp the disease name etc. with the fewest possible number of observations for each individual patient.
[0053] On the other hand, if the termination condition is met (S26; YES), the process shown in Figure 5 ends.
[0054] In addition, the observation target selection device 3 repeatedly predicts labels for data containing the specific observation target selected by the prediction unit 34 in the above-described processes (S22, S23), but is not limited to this. For example, the observation target selection device 3 may, after selecting all observation targets in process (S22), predict labels for data containing all observation targets selected in process (S22) in process (S23).
[0055] With this, the processing or operation of the prediction phase is completed.
[0056] [Evaluation Results] Next, we will explain the evaluation results of the predictions made by the observation target selection device of this embodiment. Figure 6 shows the evaluation results.
[0057] Figure 6 shows the results of evaluating this embodiment using handwritten digit data. RL represents a reinforcement learning-based method, FI represents a feature estimation-based method, Var represents a method that selects the feature with the maximum variance, and Random represents a method that randomly selects features. As shown in Figure 6, this embodiment achieves a higher accuracy rate compared to other methods.
[0058] [Main effects of the embodiment] As described above, according to this embodiment, the observation target selection device 3 has the effect of being able to select a specific observation target indicated by a missing value, according to the individual data.
[0059] Furthermore, the observation target selection device 3, through the process described above (S24), does not require observation of all observation targets, thus reducing the cost of observation and alleviating the burden on patients in observing examination items and other observation targets.
[0060] 〔supplement〕 The present invention is not limited to the embodiments described above, and may also have the following configurations or processes (operations).
[0061] The observation target selection device 3 can be implemented using a computer and a program, and this program can be recorded on a (non-temporary) recording medium or provided via a communication network 100. [Explanation of Symbols]
[0062] 1. Communication System 3. Observation target selection device 5. Communication terminals 31 Input section 32 Predictive Model Learning Section 33 Selection Model Learning Unit 34 Prediction Section 39 Output section
Claims
1. An observation target selection device for selecting an observation target to be observed in the prediction phase, An input unit for inputting data that includes predetermined missing values due to unobserved data, A selection model that includes data showing the characteristics of observed objects, and a prediction unit that uses a selection model that includes data showing An output unit that outputs selection result data indicating the specific observation target selected by the prediction unit, It has, The prediction unit predicts labels based on the data containing predetermined missing values input by the input unit, using a machine learning-prepared prediction model for predicting labels for data containing predetermined missing values. The output unit outputs prediction result data showing the label predicted by the prediction unit. The prediction unit then predicts the label again based on the updated data, which is obtained by reflecting the characteristics of the specific observation target obtained from new observations based on the selected data into the predetermined missing values. The output unit outputs prediction result data showing the label predicted again by the prediction unit. Observation target selection device.
2. A method for selecting an object to be observed in the prediction phase, which is performed by an object selection device for selecting an object to be observed, The aforementioned observation target selection device is Input processing involves inputting data that includes predetermined missing values due to unobserved data, A pre-trained selection model that includes data showing the characteristics of observed objects, and a prediction process that selects a specific object based on data containing specific missing values due to unobserved objects, using a pre-trained selection model for selecting a specific object from among multiple candidate objects indicated by the predetermined missing values, based on data containing specific missing values due to unobserved objects. An output process that outputs selection result data indicating the specific observation target selected by the prediction process, Execute, The prediction process includes a process of predicting labels based on the data containing the predetermined missing values input by the input process, using a machine learning-prepared prediction model for predicting labels for data containing the predetermined missing values, The output process includes a process that outputs prediction result data showing the labels predicted by the prediction process, The prediction process includes a process of predicting the label again based on updated data, which is obtained by reflecting the characteristics of the specific observed object obtained from new observations based on the selected data into the predetermined missing values. The output process includes a process to output prediction result data showing the label predicted again by the prediction process. Method for selecting subjects for observation.
3. A program that causes a computer to perform the method described in claim 2.