A random forest-based personnel risk analysis method

Abstract

The invention discloses a random forest-based personnel risk analysis method, which relates to the field of machine learning and comprises a target category selection process, a training process and an evaluation process. Comprising the following steps: selecting YD related to a danger degree from personnel multi-dimensional data as a target category; setting a danger degree index PYD, and establishing a training data set PreTrain Set set; training a random forest regression model to obtain RFM, GPYD and LNPY; calculating to obtain that the risk of the evaluated personnel is a multiple SPYD of the average level, so that the relative dangerousness can be visually evaluated. According to the evaluation method, the training data do not contain subjective judgment or scale test results of people, the model quality is higher, meanwhile, the data labeling cost is reduced, the model updating speed is increased, and the model can adapt to environment changes more quickly and at lower cost.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to a random forest-based personnel risk analysis method. Background technique [0002] At present, there are mainly three types of personnel risk assessment: one is empirical assessment. Due to the limitation of the conditions at the time, the risk assessment method is mainly defined by the evaluator’s intuitive feeling of the transformation performance. This type of method relies on the evaluator’s subjective judgment. The accuracy rate is not high, and the key to the evaluation process is to rely on the quality of the evaluator. For the same person to be evaluated, the conclusions of different evaluators may be quite different; The accuracy rate no longer depends on the subjective judgment of the evaluator, but the preparation of mature and effective scales takes a very long time, and most scales cannot be treated differently with changes in social environment, geographical region,...

AI Technical Summary

Problems solved by technology

[0002] At present, there are mainly three types of personnel risk assessment: one is empirical assessment. Due to the limitation of the conditions at the time, the risk assessment method is mainly defined by the evaluator’s intuitive feeling of the transformation performance. This type of method relies on the evaluator’s subjective judgment. The accuracy rate is not high, and the key to the evaluation process is to rely on the quality of the evaluator. For the same person to be evaluated, the conclusions of different evaluators may be quite different; The accuracy rate no longer depends on the subjective judgment of the evaluator, but the preparation of mature and effective scales takes a very long time, and most scales cannot be treated differently with changes in social environment, geographical region, cultural customs, and causes of formation; the third The first is a risk assessment method driven by machine learning data, relying on multi-dimensional data accumulated over the years, using machine learning algorithms to train classification models, and then classifying the evaluators through the trained model

[0003] The risk assessment method based on machine learning is still in the exploratory stage in general, but in the actual operation process, there are generally problems of difficult operation and low evaluation reliability and validity

Existing methods are all based on experience, mostly in the form of classification, that is, to divide the personnel to be evaluated into two categories of dangerous and non-dangerous, or into a few categories such as high-risk, medium-risk, and low-risk, lacking quantitative indicators of risk levels , it is impossible to accurately distinguish the different risk levels of different persons to be evaluated in a category; secondly, the labeling method of the training data set depends on subjective judgment or scale test results, not only the cost of labeling is high, but both subjective judgment and scale test as mentioned above There is a large error, resulting in low accuracy of the training data set, and the model trained by the error training data is bound to be inaccurate

Although some methods can obtain quantitative indicators of risk levels, it is difficult to explain the meaning of their values, and it is difficult for users to understand and grasp them.

Images