A Prediction Method for Maritime Accident Types Based on Selective Ensemble Learning

Abstract

The present invention provides a method for predicting maritime accident types based on selective ensemble learning. In view of the characteristics of strong timeliness of maritime accident data and many involved risk factors, and considering the accuracy and universality required by the maritime accident type prediction model, a two-stage model selection method is proposed by using the selective ensemble learning technology. This method takes into account both model accuracy and diversity, constructs a maritime accident type prediction model by selecting a group of individual learners with excellent performance, improves the generalization ability and prediction accuracy of the model, realizes the rapid and reliable prediction of maritime accident types, and enhances the pertinence of maritime accident prevention.

Description

Technical Field

[0001] The present invention relates to the field of prediction of maritime accident types, and specifically relates to a method for predicting maritime accident types based on selective ensemble learning. Background Art

[0002] With the rapid development of economic globalization, shipping plays an important role in international trade, and maritime safety has always been an issue of concern in the international shipping industry. However, maritime transportation is affected by complex and uncertain risk factors, which are extremely likely to lead to the occurrence of maritime accidents, causing serious economic losses, casualties, and marine pollution. In order to prevent the occurrence of maritime accidents, it is necessary to accurately and reliably predict the types of maritime accidents. Predicting the types of maritime accidents infers the possible types of future maritime accidents based on historical accident data, which helps to improve the maritime accident early warning mechanism and proactively prevent the occurrence of maritime accidents, and is of great significance for improving the safety level of maritime transportation.

[0003] Machine learning technology provides an effective method for solving multi-variable, non-linear, and complex problems, and is widely applied in multiple fields of risk prediction.

[0004] At present, although a single machine learning method can analyze the multi-factor coupling problems involved in maritime accidents, due to the strong timeliness of maritime big data and the dynamic uncertainty characteristics of the risk factors involved, a single machine learning method cannot adapt to the rapidly changing new maritime data and the model has poor robustness. Summary of the Invention

[0005] The present invention proposes a method for predicting maritime accident types based on selective ensemble learning to fill the gap in selective ensemble learning technology in maritime accident prediction and further improve the accuracy and generalization ability of the maritime accident type prediction model.

[0006] To this end, the present invention provides the following technical solutions:

[0007] The present invention provides a method for predicting maritime accident types based on selective ensemble learning, including the following steps:

[0008] Sort out maritime accident investigation reports and establish a maritime accident dataset;

[0009] Perform data preprocessing on the maritime accident dataset and randomly divide the dataset into a training set and a test set;

[0010] Determine the individual learner algorithm, optimize the hyperparameters on the training set, establish multiple individual learners, and form an individual learner set;

[0011] A two-stage model selection method is adopted to select the pre-set individual learners in advance that can optimize the prediction performance of the integrated model for ensemble learning; the two-stage model selection method includes: a model accuracy selection stage and a model diversity selection stage;

[0012] Use the stacking combination strategy for the pre-set individual learners of ensemble learning to construct a maritime accident type prediction model, train the integrated model on the training set, and evaluate the model performance on the test set;

[0013] Use the trained integrated model to predict the maritime accident type.

[0014] Furthermore, the model accuracy selection stage includes:

[0015] Remove the individual learners with a prediction accuracy lower than the preset value on the test set from the set of individual learners;

[0016] The model diversity selection stage includes:

[0017] Select the first individual learner from the remaining individual learners in the set of individual learners. The first individual learner is the individual learner with the best prediction performance; and add the first individual learner to the subset of individual learners of the integrated model. The individual learner with the best prediction performance is the individual learner with the highest sum of multiple evaluation indicators;

[0018] Select the second individual learner from the remaining individual learners in the set of individual learners. The second individual learner is the individual learner with the greatest difference from the first individual learner, add the second individual learner to the subset of individual learners, and calculate the prediction performance of the ensemble learning model constructed based on the subset of individual learners;

[0019] Select the third individual learner from the remaining individual learners in the set of individual learners. The third individual learner is the individual learner with the greatest difference from the integrated model constructed by the first individual learner and the second individual learner based on the bagging combination strategy, add the third individual learner to the subset of individual learners, and calculate the prediction performance of the ensemble learning model constructed based on the subset of individual learners; the model difference is the difference degree of the classification results obtained by different models on the same test set;

[0020] Repeat the steps of the model diversity selection stage until the set of individual learners is rearranged into a new sequence in the subset of individual learners.

[0021] Furthermore, the maritime accident dataset includes accident type, accident time, ship attributes, and accident risk factors.

[0022] Further, data preprocessing is performed on the maritime accident dataset, including:

[0023] Converting continuous variables into categorical variables;

[0024] Deleting variables with a frequency less than 5%.

[0025] Further, hyperparameter optimization uses random search with 30 iterations and 10-fold cross-validation to determine the set of hyperparameters with the best model accuracy.

[0026] Further, the evaluation metrics include: accuracy, precision, recall, and F1-score.

[0027] Further, the diversity of the ensemble model is determined using two diversity measurement methods: Q-statistic and double-failure metric.

[0028] Further, for two different learners L i and L j The calculation formula for the Q-statistic is as follows:

[0029]

[0030] For two different learners L i and L j The calculation formula for the double-failure metric is as follows:

[0031]

[0032] Where N 11 is the number of samples correctly classified by L i and L j N 00 is the number of samples misclassified by the two learners L i and L j N 10 represents the number of samples that are correctly classified by L i but misclassified by L j N 01 represents the number of samples that are correctly classified by L j but misclassified by L i N represents the total number of samples.

[0033] The beneficial effects of the present invention are:

[0034] (1), The present invention adopts ensemble learning technology. Compared with traditional single machine learning technology, it can reduce the risk of falling into local optimal solutions and misclassification, thereby improving the accuracy and generalization ability of the prediction model.

[0035] (2) The present invention proposes a two-stage model selection method, introduces selective ensemble learning technology, and deletes redundant individual learners, providing a new way for quickly, accurately, and reliably predicting the types of maritime accidents.

[0036] (3) The present invention utilizes selective ensemble learning technology to mine maritime accident data, which can quickly predict the possible types of maritime accidents and contribute to the active prevention of maritime accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 It is a flowchart of the method for predicting the types of maritime accidents based on selective ensemble learning of the present invention;

[0039] Figure 2 It is a bar chart of the historical dataset of maritime accidents in the embodiments of the present invention;

[0040] Figure 3 It is a schematic diagram showing the influence of the number of individual learners on the prediction performance of the ensemble model in the embodiments of the present invention;

[0041] Figure 4 It is a schematic diagram of the results of the model performance in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0043] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0044] Ensemble learning is one of the most important research directions in machine learning. Bagging, Boosting, and Stacking are three classic ensemble learning methods. By combining multiple individual learners, the prediction performance and generalization ability of the model can be improved, and the computational error caused by a single machine learning model can be reduced. Although the ensemble learning model has advantages in risk prediction, with the increase in the number of individual learners, the prediction speed of the ensemble model decreases significantly and the storage space increases. Therefore, the selective ensemble learning method emerges as the times require. Selective ensemble learning constructs an ensemble model by removing poorly performing individual learners and only selecting a group of highly diverse and accurate individual learners, in order to obtain better prediction performance and improve the generalization ability of the model. As a new machine learning technology with excellent performance and broad prospects, selective ensemble learning has not been studied in the field of maritime accident prediction. Based on this, the present invention provides a method for predicting maritime accident types based on selective ensemble learning to adapt to rapidly changing new maritime data and improve the generalization ability of the model.

[0045] As Figure 1 shown, a method for predicting maritime accident types based on selective ensemble learning provided by the present invention includes the following steps:

[0046] S1. Sort out maritime accident investigation reports and establish a maritime accident dataset.

[0047] Among them, the maritime accident dataset includes accident types, accident times, ship attributes, and accident risk factors.

[0048] The maritime accident dataset used in this embodiment is compiled from maritime accident investigation reports published on the official websites of 8 maritime investigation agencies, including 492 maritime accidents that occurred from 2011 to 2020. As Figure 2As shown. This dataset contains the five most common types of maritime accidents, namely collision, grounding / stranding, contact, sinking, and fire / explosion. The original data contains a total of 78 variables, covering accident types, accident times, ship types, and unsafe behaviors, etc. Some variables in the original maritime accident dataset are shown in Table 1.

[0049] Table 1

[0050]

[0051]

[0052] S2. Perform data preprocessing on the maritime accident dataset and randomly divide the dataset into a training set and a test set.

[0053] In this embodiment, data preprocessing is performed on the maritime accident dataset, and continuous variables such as time, gross tonnage, ship length, and ship age are transformed into categorical variables. Then, this embodiment deletes the variables with a frequency less than 5% in the dataset, and the variables of the maritime accident data are reduced to 39. The dataset is randomly divided into a training set (80%) and a test set (20%).

[0054] S3. Determine the individual learner algorithm, optimize the hyperparameters on the training set to develop multiple individual learners with good prediction performance, and form the individual learner set L.

[0055] Among them, when determining the individual learner algorithm, the prediction performance of the algorithm and the differences between algorithms need to be considered, and heterogeneous individual learner algorithms are selected.

[0056] In this embodiment, 7 heterogeneous machine learning algorithms are determined, namely polynomial logistic regression (MLR), feedforward neural network (FNN), support vector machine (SVM), K-nearest neighbor (KNN), classification and regression tree (CART), random forest (RF), and extreme gradient boosting algorithm (XGBoost). This embodiment uses the random search method to optimize the hyperparameters through 10-fold cross-validation and 30 iterations. The process is as follows: The training data is divided into 10 subsets, and each subset is sequentially selected as the validation set, and the remaining 9 subsets are combined into the training set to obtain 10 training sets and validation sets. On this basis, hyperparameter optimization constructs 10 models with different hyperparameter combinations using 30 iterations, and calculates the average accuracy of these 10 models as the model optimization criterion. Search for the hyperparameter set with the highest average accuracy as the optimal hyperparameters to construct individual learners with good prediction performance.

[0057] Finally, the individual learner set L = {MLR, FNN, SVM, KNN, CART, RF, XGBoost}.

[0058] S4. Remove the individual learners in L whose prediction accuracy on the test set is lower than 75%.

[0059] In this embodiment, the accuracies of the 7 individual learners in the individual learner set L on the test set are calculated respectively. Among them, the accuracies of FNN (70.71%) and XGBoost (74.75%) are lower than 75% and are removed from L to avoid the negative impact brought by the poorly performing individual learners. At this time, L = {MLR, SVM, KNN, CART, RF}.

[0060] S5. Select the first individual learner from the remaining individual learners in L and add it to the subset E of individual learners of the ensemble model i as the initial individual learner E1.

[0061] Among them, the first individual learner is the individual learner with the best prediction performance. To select the individual learner with the best prediction performance, four evaluation metrics, namely accuracy, precision, recall, and F1-score, are required. The individual learner with the best prediction performance is the individual learner with the highest sum of multiple evaluation metrics.

[0062] To understand the basic meaning of the metrics used, Table 2 provides an example of a confusion matrix for a three-class classification. The columns and rows in the matrix represent the predicted class and the observed class respectively. For example, N 11 represents the number of data where the actual class is 1 and the predicted result is also class 1, N 12 represents the number of data where the actual class is 1 and the predicted result is class 2, N 13 represents the number of data where the actual class is 1 and the predicted result is class 3, N 21 represents the number of data where the actual class is 2 and the predicted result is class 1; N 22 represents the number of data where the actual class is 2 and the predicted result is class 2; N 23 represents the number of data where the actual class is 2 and the predicted result is class 3; N 31 represents the number of data where the actual class is 3 and the predicted result is class 1; N 32 represents the number of data where the actual class is 3 and the predicted result is class 2; N 33 represents the number of data where the actual class is 3 and the predicted result is class 3.

[0063] Table 2

[0064]

[0065] Accuracy refers to the ratio of the correctly predicted observations to the total observations (TN), and the calculation formula is as follows:

[0066]

[0067] Precision refers to the ratio of the correctly predicted observations in a specific category to all the predicted values in that category, and the calculation formula is as follows:

[0068]

[0069]

[0070] Recall refers to the ratio of the correctly predicted observations under a specific category to all the actual observations in that category, and the calculation formula is as follows:

[0071]

[0072]

[0073] The F1-score value is an index calculated based on precision and recall, which is used to represent the generalization ability of the model. The calculation formula is as follows:

[0074]

[0075] In this embodiment, four evaluation indicators, namely accuracy, precision, recall, and F1-score, are used to evaluate the performance of the remaining 5 individual learners in L. The results are shown in Table 3. It can be seen from Table 3 that RF has the highest prediction accuracy, but compared with SVM, RF performs poorly in terms of F1-score. Considering the results of the four evaluation indicators comprehensively, this embodiment selects the SVM model with an accuracy of 83.84% and an F1-score of 76.14% on the test set as the initial individual learner E1. At this time, L = {MLR, KNN, CART, RF}, E i = {SVM}.

[0076] Table 3

[0077]

[0078] [[ID= forty]]S6. Select the second individual learner from the remaining individual learners in L. The second individual learner is the individual learner E2 with the largest difference from E1, and add it to E i , and calculate the prediction performance of the ensemble learning model constructed based on the individual learners in E i at this time.

[0079] In this embodiment, the pairwise differences between the remaining 4 individual learners in L and the SVM model are calculated respectively, and the Q statistic and the double failure metric (DF) are used to determine that there is the greatest degree of difference between the CART model and the SVM model. Therefore, CART is added to E i and recorded as E2. At this time, L = {MLR, KNN, RF}, E i = {SVM, CART}.

[0080] S7. Select the third individual learner from the remaining individual learners in L. The third individual learner is the individual learner E3 with the greatest difference from the ensemble model constructed by E1 and E2 based on the bagging combination strategy, and add it to E i , and calculate the prediction performance of the ensemble learning model constructed by the individual learners in E i at this time.

[0081] In this embodiment, CART and SVM are bagging integrated, and then the Q statistic values and DF values between the remaining 3 individual learners in L and this integrated model are calculated. The results show that KNN has the greatest degree of difference from this integrated model. Therefore, KNN is added to E i and recorded as E3. At this time, L = {MLR, RF}, E i = {SVM, CART, KNN}.

[0082] Suppose there are L learners, L i and L j (i, j = 1, 2,..., L, i ≠ j). For two different learners, N 11 (N 00 ) is the number of samples correctly (incorrectly) classified by the two learners L i and L j , and N 10 (N 01 ) represents the number of samples that are correctly classified by L i (L j ) but incorrectly classified by L j (L i ), as shown in Table 4.

[0083] Table 4

[0084]

[0085]

[0086] The Q statistic calculation formula for two learners L i and L j is as follows:

[0087]

[0088] Two learners L i and L j The formula for the double failure metric (DF) is as follows:

[0089]

[0090] S8. Repeat this process until L is rearranged into a new sequence in E i and form a new sequence.

[0091] In this embodiment, the above process is repeated until L is rearranged into a new sequence in E i and form a new sequence. During this process, the Q statistic values and DF values between individual learners can be seen in Table 5. Finally, E i ={SVM, CART, KNN, RF, MLR}.

[0092] Table 5

[0093]

[0094] The above steps S4 - S8 are the model selection process. Model selection mainly includes two stages: model accuracy (S4) and model diversity (S5 - S8).

[0095] S9. Select the top pre - set number of individual learners that can make the prediction performance of the ensemble model reach the optimal, and perform ensemble learning. Adopt the stacking combination strategy to construct a maritime accident type prediction model, train the ensemble model on the training set, and evaluate the model performance on the test set.

[0096] In this embodiment, according to the order of individual learners in E i ={SVM, CART, KNN, RF, MLR}, calculate the influence of the number of individual learners on the ensemble model respectively. The results are as Figure 3 shown. When selecting the first 2 individual learners (SVM and CART) for combination, the accuracy of the stacking - based ensemble model is 88.89%, and the F1 - score is 82.35%. Compared with the optimal individual learner SVM, the accuracy of the selective ensemble model is increased by 5.05%, and the F1 - score is increased by 6.21%, indicating that combining multiple individual learners can improve the prediction performance of the model. However, when combining 5 individual learners, the prediction performance of the model is not satisfactory, with an accuracy of 83.84% and an F1 - score of 73.61%, which reveals the over - fitting problem caused by too many individual learners. The poor performance of some individual learners affects the reliability of the ensemble model.

[0097] Therefore, in engineering applications, the integration of the first two individual learners will yield more accurate and reliable prediction results for the types of maritime accidents. This indicates that the method proposed in the present invention not only eliminates a large number of redundant individual learners, thus significantly reducing the integration scale, but also further improves the performance of predicting the types of maritime accidents.

[0098] To better evaluate the prediction performance of the method proposed in the present invention on the test set, Table 6 provides the confusion matrix results of this selective ensemble learning model. As can be seen from Table 6, there are no misclassified collision accident records, which means that the model can effectively predict collision accidents. This result is also reflected in Figure 4 where the prediction precision rate of collision accidents reaches 100%, followed by sinking accidents (85.71%) and stranding accidents (81.25%). However, the prediction precision rate of fire accidents is relatively poor, at 63.64%, while the recall rate of fire accidents reaches 100%, indicating that all fire accidents have been successfully predicted. It should be noted that the data used in the study was manually sorted, and there may be subjective problems in this process, which may lead to misclassification.

[0099] Table 6

[0100]

[0101] The beneficial effects of the above embodiments are as follows:

[0102] (1) In the embodiments of the present invention, the ensemble learning technology is adopted. Compared with the traditional single machine learning technology, it can reduce the risks of falling into local optimal solutions and misclassification, thereby improving the accuracy and generalization ability of the prediction model.

[0103] (2) A two-stage model selection method is proposed in the embodiments of the present invention. By introducing the selective ensemble learning technology and deleting redundant individual learners, it provides a new way for quickly, accurately, and reliably predicting the types of maritime accidents.

[0104] (3) By using the selective ensemble learning technology to mine maritime accident data in the embodiments of the present invention, it can quickly predict the possible types of maritime accidents, which helps to actively prevent the occurrence of maritime accidents.

[0105] In several embodiments provided by this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in electrical or other forms.

[0106] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0107] In addition, in each embodiment of the present invention, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0108] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0109] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of each embodiment of the present invention.

Claims

1. A method for predicting types of maritime accidents based on selective ensemble learning, characterized in that, Including the following steps: Sort out the marine accident investigation reports and establish a marine accident dataset; the marine accident dataset includes accident types, accident times, ship attributes, and accident risk factors; Perform data preprocessing on the marine accident dataset and randomly divide the dataset into a training set and a test set; Determine the individual learner algorithm, optimize the hyperparameters on the training set, establish multiple individual learners, and form an individual learner set; Adopt a two-stage model selection method to select the top preset individual learners that make the prediction performance of the ensemble model reach the optimal for ensemble learning; The two-stage model selection method includes: a model accuracy selection stage and a model diversity selection stage; Adopt a stacking combination strategy for the top preset individual learners of the ensemble learning to construct a marine accident type prediction model, train the ensemble model on the training set, and evaluate the model performance on the test set; Use the trained ensemble model to predict the marine accident types; Among them, the model accuracy selection stage includes: Remove the individual learners in the individual learner set whose prediction accuracy on the test set is lower than the preset value; The model diversity selection stage includes: Select the first individual learner from the remaining individual learners in the individual learner set, and the first individual learner is the individual learner with the best prediction performance; and add the first individual learner to the individual learner subset of the ensemble model; the individual learner with the best prediction performance is the individual learner with the highest sum of multiple evaluation indicators; Select the second individual learner from the remaining individual learners in the individual learner set, and the second individual learner is the individual learner with the greatest difference from the first individual learner, add the second individual learner to the individual learner subset, and calculate the prediction performance of the ensemble learning model constructed based on the individual learner subset; Select the third individual learner from the remaining individual learners in the individual learner set, and the third individual learner is the individual learner with the greatest difference from the ensemble model constructed by the first individual learner and the second individual learner based on the bagging combination strategy, add the third individual learner to the individual learner subset, and calculate the prediction performance of the ensemble learning model constructed based on the individual learner subset; the model difference is the difference degree of the classification results obtained by different models on the same test set; Repeat the steps of the model diversity selection stage until the individual learner set is rearranged into a new sequence in the individual learner subset; 2. The method for predicting offshore accident types based on selective ensemble learning according to claim 1, wherein, Performing data preprocessing on the marine accident dataset includes: Convert continuous variables into categorical variables; Delete variables with a frequency less than 5%; 3. A method for predicting types of maritime accidents based on selective ensemble learning according to claim 1, characterized in that Hyperparameter optimization uses 30 iterations and 10-fold cross-validation for random search to determine the set of hyperparameters with the best model accuracy; 4. A method for predicting types of maritime accidents based on selective ensemble learning according to claim 1, characterized in that The evaluation indicators include: accuracy, precision, recall, and F1-score; 5. A method for predicting offshore accident types based on selective ensemble learning according to claim 1, characterized in that, The difference of the ensemble model is determined by two diversity measurement methods: Q statistic and double failure metric.

6. A method for predicting types of maritime accidents based on selective ensemble learning according to claim 5, characterized in that, Two different learners and The Q-statistic calculation formula is as follows: ； Two different learners and The double failure metric calculation formula is as follows: ； Among them, is and the number of correctly classified samples, is the number of samples misclassified by two learners and ; represents the number of samples that are correctly classified but misclassified; represents the number of samples that are correctly classified but misclassified; represents the total number of samples.

Images