Power enterprise personnel competency assessment method based on DL and SVM
Through the hybrid evaluation model of DL and SVM, the problems of manual dependence and low data utilization in the competency assessment of traditional power enterprise personnel are solved, efficient and accurate competency assessment and full-dimensional profiling are achieved, and the efficiency of enterprise management is improved.
Patent Information
- Application Number
- CN202510712802.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2025-09-19
AI Technical Summary
Traditional power enterprise personnel competency assessment methods rely on manual experience, have low data utilization, poor dynamic adaptability, and are difficult to accurately capture complex relationships and hidden patterns.
A hybrid evaluation model based on deep learning (DL) and support vector machine (SVM) is adopted to construct a competency hybrid model through multimodal feature extraction and dynamic weight allocation, combined with transfer learning and small sample SVM training.
It improves the accuracy, efficiency and interpretability of assessments, realizes full-dimensional personnel portraits, and supports efficient human resource management in enterprises.
Smart Images
Figure CN120672190A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data analysis and artificial intelligence application in the power industry, and specifically to a competency assessment method for power enterprise personnel based on DL and SVM. Background Art
[0002] With the rapid development of the power industry, power companies have accumulated a vast amount of personnel data, including multi-dimensional information such as basic employee information, work performance, skill levels, training history, and attendance records. Effectively utilizing this data to accurately classify power industry personnel and assess their competency is crucial for improving enterprise management efficiency, optimizing job allocation, and enhancing employee capabilities. Traditional classification methods often rely on manual experience and simple statistical analysis, making it difficult to accurately capture complex relationships and hidden patterns in the data. Deep learning (DL) can capture complex nonlinear relationships in data through multi-layer neural networks. It excels at automatically extracting high-dimensional features from raw data and can process both structured and unstructured data. Support vector machines (SVMs), as a powerful supervised learning model, excel in classification problems, with clear decision boundaries and intuitive support vectors. They are particularly well-suited for processing high-dimensional data and nonlinear relationships, and are therefore widely used in data classification tasks across various fields. Summary of the Invention
[0003] This paper focuses on the intelligent upgrade of the competency assessment of power enterprise personnel, and provides a competency assessment method for power enterprise personnel based on DL and SVM. By integrating a hybrid assessment model with deep learning (DL) and support vector machine (SVM), it aims to solve the problems of traditional assessment methods relying on manual experience, low data utilization, and poor dynamic adaptability.
[0004] A competency assessment method for power enterprise personnel based on DL and SVM, including:
[0005] Step 1: Determine the data source: Collect multi-dimensional data of power company personnel, including basic information, work experience, educational background, training records, and behavioral event records;
[0006] Step 2: Data preprocessing: Clean the structured data in the multidimensional data collected in step 1, handle missing values and outliers, and standardize or normalize the numerical features; use natural language processing (NLP) technology to extract text features from the unstructured data in the multidimensional data; and finally obtain the original data set P (p1, p2, ..., pn), where n is the actual amount of data with normal values that can be collected in the sample;
[0007] Step 3: Determine target data: Based on expert definitions, determine the eight-dimensional index including "trend index, vitality index, experience index, awareness index, ability index, learning index, skill index, and professional index";
[0008] Step 4: DL model selection: Select a suitable DL model from deep learning models based on the characteristics of the data and evaluation requirements;
[0009] Step 5: DL model pre-training: Using transfer learning, the DL model selected in step 4 is pre-trained using a large-scale public dataset, and then fine-tuned using the power company's internal data to reduce training time and computing resource consumption;
[0010] Step 6: Feature extraction: Use the DL model pre-trained in step 5 to extract the deep features F of personnel competency from the original data set P(p1,p2,...,pn);
[0011] Step 7: Feature mapping: The deep features F output in step 6, including objective features and subjective features, are mapped to 8-dimensional indexes T (t1, t2, t3, t4, t5, t6, t7, t8) using a dynamic weight allocation mechanism to obtain a set of sample 8-dimensional indexes T, i.e., the personnel feature set T 集合 ={Ti|i∈number of samples};
[0012] Step 8: Train the small sample SVM model to obtain the initial version of the SVM model;
[0013] Step 9: Construct a competency hybrid model: Use the personnel feature set index obtained in step 7 as input and combine it with the initial version of the SVM model obtained in step 8 to construct a competency hybrid model;
[0014] Step 10: Hybrid model training: Using sample data from 500 people, randomly divide the data into training and test sets in a ratio of 8:2 to ensure balanced class distribution. Train the competency hybrid model constructed in Step 9 and modify the corresponding parameters of the DL model and the SVM model based on the output results to improve model accuracy.
[0015] Step 11: Analysis of evaluation results: Input the relevant data of power enterprise personnel into the competency hybrid model trained in step 10 to obtain the predicted results of their competency levels.
[0016] Furthermore, the data sources of the multi-dimensional data are ERP, contactless attendance and security management systems.
[0017] Furthermore, in step 3, one or more methods selected from the Delphi method, behavioral event interview method, and questionnaire survey method are used to survey human resources experts, integrate expert opinions, classify the multidimensional data collected in step 1, and substitute historical data for verification to form the 8-dimensional index.
[0018] Furthermore, the deep learning model includes a convolutional neural network (CNN) and a recurrent neural network (RNN).
[0019] Furthermore, in step 6, the trained DL model uses a multimodal Transformer structure to achieve feature fusion, and the model input-output relationship is expressed as F = Ψ(P; θ), where:
[0020] P = [p1, p2, ..., pn] ∈ R^{n × d} represents the original data matrix, n × d represents n samples and d-dimensional features;
[0021] Ψ(·) represents the pre-trained deep neural network mapping function;
[0022] θ is the set of model parameters;
[0023] F=[f1,f2,...,f m ]∈R^{m×k} is the extracted deep feature matrix, m×k represents m features and k-dimensional embedding.
[0024] Furthermore, in step 7, a dynamic weight allocation mechanism is used to map to the 8-dimensional index T(t1, t2, t3, t4, t5, t6, t7, t8), and Softmax normalization is used to map the individual feature vector to t1, t2, t3, t4, t5, t6, t7, and t8 respectively, and the value with the highest probability is obtained, which is expressed as t = Softmax(∑_{i=1}^nα_i·h_i), where:
[0025] h_i represents the importance weight of the i-th feature;
[0026] α_i=Attention(Q,K_i,V_i) is the attention weight;
[0027] n means there are n original features that need to be fused;
[0028] Traverse all features from i=1 to i=m, sum the product of attention weight α_i and importance weight h_i for each item, and select the category corresponding to the component with the largest probability as the prediction result t;
[0029] T = [t1, ..., t8] ∈ R^8 represents the DL feature vector of a data point with 8 features;
[0030] The 8-dimensional index T of each person is summarized into a person feature set T as the input of the subsequent SVM classifier.
[0031] Furthermore, in step 8, the small sample SVM model is trained to obtain the initial version of the SVM model, which specifically includes:
[0032] A small sample of 50 people was selected, and the power human resources experts obtained the 8-dimensional expert evaluation index T' (t'1, t'2, t'3, t'4, t'5, t'6, t'7, t'8) of the personnel through manual evaluation. The 8-dimensional expert evaluation index T' was standardized and the formula was:
[0033]
[0034] Where μ is the mean and σ is the standard deviation;
[0035] According to the small sample evaluation results, where competent is 1 and incompetent is 0, the radial basis kernel function is selected to build the SVM-RBF model. Grid search is used to try different C values and gamma values. The best combination is found through cross-validation to complete the SVM model selection and obtain the initial version of the SVM model.
[0036] Furthermore, step 11 also includes: using a sample of 100 people who have no job changes and no significant changes in personal abilities in recent years, and using three evaluation methods, namely, "traditional performance evaluation", "SVM model (initial)", and "hybrid model evaluation", to verify the prediction accuracy of the hybrid model.
[0037] The present invention proposes a method for evaluating the competency of power enterprise personnel based on DL and SVM, which achieves the following beneficial effects:
[0038] 1. Improve assessment accuracy:
[0039] Multimodal Deep Feature Extraction: The innovative use of DL technologies, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), combined with a multimodal Transformer architecture, automatically extracts deep features from employees' structured and unstructured multidimensional data. Unlike traditional methods of extracting features through manually designed indicators, this multimodal deep automatic extraction method captures more complex nonlinear relationships in the data, thereby more comprehensively and accurately reflecting employee competency characteristics and improving assessment accuracy. For example, in experiments, the predictive accuracy of hybrid model assessments reached 83.34%, far exceeding the 33.33% of traditional performance evaluations.
[0040] Dynamic Weight Allocation and Feature Mapping: A dynamic weight allocation mechanism is used to map extracted features into an 8-dimensional personal feature index space. The importance of different features is dynamically correlated using attention weights α_i. Compared to traditional linear correlation analysis, this approach processes multi-source data more flexibly and accurately, ensuring that assessment results better reflect the individual's actual competency and further improving assessment accuracy.
[0041] 2. Improve evaluation efficiency:
[0042] Transfer learning accelerates training: During DL model training, we employ transfer learning, pre-training the model with a large-scale public dataset and then fine-tuning it with internal data from the power company. This approach significantly reduces training time and computing resource consumption, improves model training efficiency, and rapidly provides companies with personnel competency assessment results, meeting their needs for efficient assessments.
[0043] Automated Processing: The entire evaluation process, from data collection and preprocessing to feature extraction, model training, and analysis of evaluation results, is relatively automated. This reduces manual intervention and avoids the inefficiencies inherent in traditional methods that rely heavily on manual experience and simple statistical analysis, improving overall evaluation efficiency.
[0044] 3. Enhance model generalization and interpretability:
[0045] SVM-optimized classification: We introduce SVM as a classifier, leveraging its strengths in handling high-dimensional data and nonlinear relationships. Using kernel functions (such as the RBF kernel), we construct an optimal classification hyperplane in high-dimensional space to effectively classify high-dimensional features extracted by DL. During small-sample SVM model training, we use grid search and cross-validation to determine optimal parameters. This ensures a balance between model complexity and generalization, enabling the model to better adapt to diverse personnel data and improving generalization.
[0046] Multi-source data fusion and interpretation: By combining DL with SVM, we achieve effective fusion of multi-source data, resolving the difficulties of multi-source data fusion in traditional methods. Furthermore, SVM's clear decision boundaries and intuitive support vectors make the evaluation results more interpretable, making it easier for human resources management departments to understand and apply the evaluation results and assist in decision-making.
[0047] 4. Provide a comprehensive personnel profile: By constructing eight index targets—trend index, vitality index, experience index, awareness index, ability index, learning index, skill index, and professional index—raw data is mapped onto these eight dimensions of individual characteristic index space to form a comprehensive personnel profile. Compared to traditional single-dimensional evaluation methods, this approach enables a more comprehensive assessment of the competence of power company personnel, providing a richer and more accurate reference for human resource management tasks such as job allocation and employee training. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a schematic diagram of the competency hybrid model design according to an embodiment of the present invention.
[0049] Figure 2 2 is a schematic diagram of the selection process of the initial version of the SVM model according to an embodiment of the present invention.
[0050] Figure 3 It is a schematic diagram of the implementation of the power intranet in an embodiment of the present invention.
[0051] Figure 4 2 is a schematic diagram of the personnel feature set index according to an embodiment of the present invention.
[0052] Figure 5 2. It is a schematic diagram of the results of a small sample evaluation of personnel competency according to an embodiment of the present invention.
[0053] Figure 6 2. It is a schematic diagram of competency hybrid model training according to an embodiment of the present invention. DETAILED DESCRIPTION
[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0055] See also Figure 1-6 The embodiment of the present invention provides a method for evaluating the competency of power enterprise personnel based on DL and SVM, comprising the following steps:
[0056] Step 1: Identify data sources and collect multi-dimensional data on power company personnel, including but not limited to basic information, work experience, education background, training records, behavioral event records, etc. Determine data sources: ERP, contactless attendance, security management systems, etc.
[0057] Step 2: Data preprocessing. For structured data, this involves cleaning the data (addressing missing values and outliers) and standardizing or normalizing numerical features. For unstructured data, natural language processing (NLP) techniques are used to extract text features (e.g., employee reviews and project reports). The resulting raw data set is P(p1, p2, ..., pn), where n is the amount of data with normal values that can be collected in the sample.
[0058] Step 3: Determine the target data. Based on the human resources work experience of power companies (using methods such as the Delphi method, behavioral event interview method, and questionnaire survey method to survey human resources experts, integrate expert opinions, and classify multidimensional data) and historical data for verification, eight index targets are formed: "trend index, vitality index, experience index, awareness index, ability index, learning index, skill index, and professional index." That is, the raw data collected by the human resources system needs to be mapped to the preset 8-dimensional index T (t1, t2, t3, t4, t5, t6, t7, t8) through DL technology.
[0059] Step 4: DL model selection: Based on the characteristics of the data and evaluation requirements, select a suitable deep learning model, such as convolutional neural network (CNN), recurrent neural network (RNN), etc.
[0060] Step 5: DL model pre-training, using the idea of transfer learning, first pre-train the DL model with a large-scale public dataset, and then fine-tune it with the power company's internal data to reduce training time and computing resource consumption.
[0061] Step 6: Feature Extraction: The trained DL (deep learning) model is used to extract deep-level features of personnel competencies from the original data set P (p1, p2, ..., pn). This model uses a multimodal Transformer structure to achieve feature fusion. The model input-output relationship can be expressed as F = Ψ(P; θ), where:
[0062] P = [p1, p2, ..., pn] ∈ R^{n×d} represents the original data matrix (n samples, d-dimensional features)
[0063] Ψ(·) represents the pre-trained deep neural network mapping function
[0064] θ is the set of model parameters
[0065] F=[f1,f2,...,f m ]∈R^{m×k} is the extracted deep feature matrix (m features, k-dimensional embedding)
[0066] Step 7: Feature Mapping: These output F features include objective features and subjective features, such as Figure 4Then, a dynamic weight allocation mechanism is used to map the features to an 8-dimensional index T(t1, t2, t3, t4, t5, t6, t7, t8). Softmax normalization is used to map the individual feature vectors to t1, t2, t3, t4, t5, t6, t7, and t8, respectively, and the value with the highest probability is obtained, which can be expressed as t = Softmax(∑_{i=1}^mα_i·h_i); where:
[0067] h_i represents the importance weight of the i-th feature;
[0068] α_i=Attention(Q,K_i,V_i) is the attention weight;
[0069] T = [t1, ..., t8] ∈ R^8 represents the DL feature vector of a data point with 8 features.
[0070] Summarize each person's 8-dimensional index T into the person feature set T 集合 As the input of the subsequent SVM classifier, its DL model advantages are shown in Table 1:
[0071] Table 1 Comparison between traditional analysis method and the present invention
[0072] Comparison Dimension Traditional statistical methods Solution of the present invention Feature extraction method Artificially designed indicators Multimodal depth automatic extraction Data Correlation Linear correlation analysis Attention Mechanism Dynamic Association Feature Dimension Single evaluation dimension 8-dimensional personal characteristic index space mapping
[0073] Step 8: At the same stage as step 6, a small sample SVM model training is completed. First, a small sample of 50 people is selected. Power HR experts manually evaluate the personnel to obtain the 8-dimensional index T'. Since SVM is sensitive to feature scale, the features are further standardized. The formula is: Among them, μ is the mean and σ is the standard deviation. Figure 5 The results of a small sample of personnel competency are obtained (competent is 1, incompetent is 0), the radial basis kernel function is selected, the SVM-RBF model is constructed, and the grid search (GridSearchCV) is used to confirm the optimal parameters of C (regularization parameter) and gamma (RBF kernel parameter). For example, different C values (such as 0.1, 1, 10, 100) and gamma values (such as 0.1, 1, 10) are tried. Finally, the best combination is found through cross-validation, ensuring the balance between model complexity and generalization ability. The SVM model selection is completed and the initial version of the SVM model is obtained. The selection process is as follows: Figure 2 shown.
[0074] Step 9: Construct a hybrid competency model. The design diagram is as follows: Figure 1 Its implementation and deployment in the power intranet is shown in Figure 3 As shown in Figure 2, the personnel feature set index T obtained in step 7 is used as input and combined with the initial version of the SVM model obtained in step 8 to construct a competency hybrid model.
[0075] Step 10: Hybrid model training, press Figure 6 The process shown uses sample data of 500 people to train the competency hybrid model constructed in step 9. The data is randomly divided into training and test sets in a ratio of 8:2 to ensure balanced category distribution. The DL model parameters and the corresponding parameters of the SVM model are modified according to the output results to improve accuracy.
[0076] Step 11: Analyze the evaluation results. The trained hybrid competency model can be used to evaluate the competency of power company personnel. By inputting the personnel's relevant data into the model, the predicted results of their competency level can be obtained. A sample of 100 people with no job changes and no significant changes in personal abilities between 2023 and 2024 was used to verify the model's prediction accuracy using three evaluation methods: "traditional performance evaluation," "SVM model (initial)," and "hybrid model evaluation." The experimental results are shown in Table 2:
[0077] Table 2 Experimental results
[0078]
[0079] The present invention designs a hierarchical model structure, in which the DL layer is responsible for feature extraction: high-dimensional features in unstructured data such as employee text resumes and operation records are automatically extracted through DL (such as BERT and CNN), and multimodal feature fusion is achieved by combining structured data (such as performance scores), constructing a full-dimensional personnel portrait, and breaking through the limitations of a single data type; the SVM layer outputs the classification results: SVM is introduced as a classifier, and the optimal classification hyperplane is constructed in the high-dimensional space through the kernel function (such as RBF kernel), which can effectively process the high-dimensional features extracted by DL, unify different data types, and avoid the difficulty of multi-source data fusion in traditional methods, thereby realizing automated feature extraction, precise classification and multi-source data fusion, significantly improving the accuracy, efficiency and interpretability of competency assessment of personnel in power enterprises, providing intelligent competency assessment results for human resources users of power enterprises, and assisting human resource management decision-making.
[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.
Claims
1. A method for evaluating the competency of power enterprise personnel based on DL and SVM, characterized by: The steps include: Step 1: Determine the data source: Collect multi-dimensional data of power company personnel, including basic information, work experience, educational background, training records, and behavioral event records; Step 2: Data preprocessing: Clean the structured data in the multidimensional data collected in step 1, handle missing values and outliers, and standardize or normalize the numerical features; use natural language processing (NLP) technology to extract text features from the unstructured data in the multidimensional data; and finally obtain the original data set P (p1, p2, ..., pn), where n is the actual amount of data with normal values that can be collected in the sample; Step 3: Determine target data: Based on expert definitions, determine the "trend index, vitality index, experience index, awareness index, ability index, learning index, skill index, and professional index." Define the trend index as t1, the vitality index as t2, the experience index as t3, the awareness index as t4, the ability index as t5, the learning index as t6, the skill index as t7, and the professional index as t8. Each employee's competency influencing factor is represented by the 8-dimensional index T (t1, t2, t3, t4, t5, t6, t7, t8); Step 4: DL model selection: Select a suitable DL model from deep learning models based on the characteristics of the data and evaluation requirements; Step 5: DL model pre-training: Using transfer learning, the DL model selected in step 4 is pre-trained using a large-scale public dataset, and then fine-tuned using the power company's internal data to reduce training time and computing resource consumption; Step 6: Feature extraction: Use the DL model pre-trained in step 5 to extract the deep features F of personnel competency from the original data set P(p1,p2,...,pn); Step 7: Feature mapping: The deep features F output in step 6, including objective features and subjective features, are mapped to each person’s 8-dimensional index T (t1, t2, t3, t4, t5, t6, t7, t8) using a dynamic weight allocation mechanism to obtain a set of sample 8-dimensional indexes T, i.e., the person feature set T 集合 ={Ti|i∈number of samples}; Step 8: Train the small sample SVM model to obtain the initial version of the SVM model; Step 9: Construct a competency hybrid model: transform the personnel feature set T obtained in step 7 into 集合 As input, combined with the initial version of the SVM model obtained in step 8, a competency hybrid model is constructed; Step 10: Hybrid model training: Using sample data from 500 people, randomly divide the data into training and test sets in a ratio of 8:2 to ensure balanced class distribution. Train the competency hybrid model constructed in Step 9 and modify the corresponding parameters of the DL model and the SVM model based on the output results to improve model accuracy. Step 11: Analysis of evaluation results: Input the relevant data of power enterprise personnel into the competency hybrid model trained in step 10 to obtain the predicted results of their competency levels.
2. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1 is characterized in that: The data sources of the multi-dimensional data are ERP, contactless attendance and security management systems.
3. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1 is characterized in that: In step 3, one or more methods selected from the Delphi method, behavioral event interview method, and questionnaire survey method are used to survey human resources experts, integrate expert opinions, classify the multidimensional data collected in step 1, and substitute historical data for verification to form the 8-dimensional index.
4. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1 is characterized in that: The deep learning model includes convolutional neural network (CNN) and recurrent neural network (RNN).
5. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1 is characterized in that: In step 6, the trained DL model uses a multimodal Transformer structure to achieve feature fusion. The model input-output relationship is expressed as F = Ψ(P; θ), where: P = [p1, p2, ..., pn] ∈ R^{n × d} represents the original data matrix, n × d represents n samples and d-dimensional features; Ψ(·) represents the pre-trained deep neural network mapping function; θ is the set of model parameters; F=[f1,f2,...,f m ]∈R^{m×k} is the extracted deep feature matrix, m×k represents m features and k-dimensional embedding.
6. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1, characterized in that: In step 7, a dynamic weight allocation mechanism is used to map to the 8-dimensional index T(t1, t2, t3, t4, t5, t6, t7, t8). Softmax normalization is used to map the individual feature vector to t1, t2, t3, t4, t5, t6, t7, and t8, respectively, and the value with the highest probability is obtained, which is expressed as t = Softmax(∑_{i=1}^nα_i·h_i), where: h_i represents the importance weight of the i-th feature; α_i=Attention(Q,K_i,V_i) is the attention weight; n means there are n original features that need to be fused; Traverse all features from i=1 to i=m, sum the product of attention weight α_i and importance weight h_i for each item, and select the category corresponding to the component with the largest probability as the prediction result t; T = [t1, ..., t8] ∈ R^8 represents the DL feature vector of a data point with 8 features; Summarize each person's 8-dimensional index T into the person feature set T 集合 As the input of the subsequent SVM classifier.
7. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1, characterized in that: In step 8, the small sample SVM model is trained to obtain the initial version of the SVM model, which specifically includes: A small sample of 50 people was selected, and the power human resources experts obtained the 8-dimensional expert evaluation index T' (t'1, t'2, t'3, t'4, t'5, t'6, t'7, t'8) of the personnel through manual evaluation. The 8-dimensional expert evaluation index T' was standardized and the formula was: Where μ is the mean and σ is the standard deviation; According to the small sample evaluation results, where competent is 1 and incompetent is 0, the radial basis kernel function is selected to build the SVM-RBF model. Grid search is used to try different C values and gamma values. The best combination is found through cross-validation to complete the SVM model selection and obtain the initial version of the SVM model.
8. The method for evaluating the competency of electric power enterprise personnel based on DL and SVM according to claim 1, characterized in that: Step 11 also includes: using a sample of 100 people who have not changed their positions and have not had significant changes in their personal abilities in recent years, and using three evaluation methods: "traditional performance evaluation", "SVM model (initial)", and "hybrid model evaluation" to verify the prediction accuracy of the hybrid model.