Suitable Deep Learning Classifier Recommendation for Multi-variate Time Series Classification

Suitable Deep Learning Classifier Recommendation for Multi-variate Time Series Classification | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Suitable Deep Learning Classifier Recommendation for Multi-variate Time Series Classification Rui Gan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3999075/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Time series classification can be categorized into two main types: univariate and multivariate. The major difference between the two is that multivariate datasets have multiple dimensions, which makes them more intricate than univariate datasets. Selecting a suitable classifier for multivariate time series classification can be challenging, as it involves trying out several algorithms to handle the intricate data. While some studies have been conducted on automatically finding suitable classifiers for univariate datasets, there are currently no such studies for multivariate ones. To fill this gap, this paper introduces a recommendation system for deep-learning classifiers that are suitable for multivariate time series classification. The experiments demonstrate that this recommendation system can automatically identify the suitable deep-learning classifier for some multivariate time series datasets. Multivariate Time Series Classification Metal-Learning Deep-Learning Figures Figure 1 Figure 2 1. Introduction Time series classification is a highly versatile tool that has been applied in various fields, including bioinformatics and finance. Since each domain has its own distinct time series data format, the classification problem can be categorized as univariate time series classification or multivariate time series classification. As the "No Free Lunch" theorem[ 1 ] suggests, there's no one algorithm that can provide the best solution for all problems. Therefore, discovering a suitable classifier for a particular time series classification problem necessitates a trial-and-error approach. Is there a method to streamline this process and make finding a suitable classifier simpler? Yes, that is correct. A study by Abanda.A, published in [ 2 ], a time series classifier recommendation (TSCR) system was developed using a meta-learning approach to identify the suitable classifier for univariate time series datasets. One of the primary challenges for the TSCR system was generating a meta-feature dataset, known as landmarker, in a timely fashion. Once this landmarker was acquired, a meta-learner could be trained. The author of [ 2 ] utilized 112 subsample univariate time datasets from the UCR repository [ 3 ] to train 24 classifiers and determine the classification accuracy of each one. These calculations were treated as attributes of each instance in the meta-feature dataset, resulting in a meta-feature dataset size of 112*24, with 112 denoting the number of time series datasets and 24 denoting the number of classifiers. The outcomes of extensive experimentation on 24 classifier algorithms publicly available in [ 4 ] were then utilized to label each instance in the meta-feature dataset. Therefore, the issue of classifier recommendation can be seen as a multi-label learning problem, with each label indicating the likelihood of a classifier being well-suited for a specific dataset. At present, the TSCR system is the only thoroughly researched option for suggesting time series classifiers. However, it is limited to univariate time series. Furthermore, out of the 24 candidate classifiers in [ 2 ], only 2 are deep-learning classifiers. Multivariate time series datasets are inherently more complex due to their multiple dimensions, making them more intricate than univariate datasets. Additionally, recent research has introduced new deep-learning algorithms for solving time series classification problems. In this article, I will introduce a multivariate time series classification recommendation (MTSCR) system called DL-MTSCR with a meta-learning approach. This system can automatically identify the suitable deep-learning classifier for multivariate time series classification. For those interested, the source code can be downloaded from https://gitee.com/kengan1013/esdl-mtsf . The DL-MTSCR system faces a crucial hurdle in constructing a meta-feature dataset for complex data. To overcome this challenge, the system combines two efficient methods that simplify multivariate time series data and facilitate the creation of a meta-feature dataset. The rest of the paper is organized as follows: Section 2 presents the framework for the recommendation system while highlighting the core idea of rapidly constructing a meta-feature dataset for multivariate time series data. Section 3 showcases the results of experimentation, and finally, Section 4 provides concluding remarks and suggests possible directions for future research. 2. Deep-Learning Classifier Recommendation in DL-MTSCR The difference between univariate time series and multivariate time series lies in their respective number of dimensions. In an univariate time series dataset, each instance has a dimension of 1, while a multivariate time series dataset has a dimension greater than 1. As such, a recommendation system for multivariate time series classifiers can be developed in the following manner: To begin, giving a set of multivariate time series datasets D = { \({D}_{1}\) , \({D}_{2}\) … \({D}_{i}\) }. Each dataset \({D}_{i}\) contains a set of n instance label pairs, represented by \({D}_{i}\) = {( \({X}_{1}\) , \({y}_{1}\) ), ( \({X}_{2}\) , \({y}_{2}\) ). .. ( \({X}_{n}\) , \({y}_{n}\) )}, where \({y}_{n}\) is the label associated to \({X}_{n}\) . Each \({X}_{n}\) is a list of d vectors, where \({X}_{n}\) = [ \({x}_{1}\) , \({x}_{2}\) · · · \({x}_{d}\) ]. Each vector \({x}_{d}\) is a time series data vector with m feature value, represented by \({x}_{d}\) = [ \({f}_{1}\) , \({f}_{2}\) … \({f}_{m}\) ]. Therefore, \({X}_{n}\) can be viewed as a matrix with d rows and m columns, where d is the number of time series and m is the length of each time series. Additionally, there is a set of deep-learning classifiers C= { \({C}_{1}\) , \({C}_{2}\) … \({C}_{j}\) }. Next, the recommendation system utilizes all multivariate time series datasets in D to construct a meta-feature dataset M = {( \({M}_{1}\) , \({Y}_{1}\) ), ( \({M}_{2}\) , \({Y}_{2}\) )…( \({M}_{i}\) , \({Y}_{i}\) )}, where \({M}_{i}\) is a feature vector pertaining to dataset \({D}_{i}\) . Therefore, \({M}_{i}\) = [ \({m}_{i,1}\) , \({m}_{i,2}\) … \({m}_{i,j}\) ] where each \({m}_{i,j}\) is a meta-attribute value about dataset \({ D}_{i}\) and deep-learning classifier \({C}_{j}\) . Additionally, each \({M}_{i}\) also has a label vector \({Y}_{i}\) , each feature value in \({Y}_{i}\) corresponds to the classification accuracy of each classifier \({C}_{j}\) on dataset \({D}_{i}\) . Consequently, the labeled meta-feature dataset can be expressed as \({M}_{i}\) = {[ \({m}_{i,1}\) , \({m}_{i,2}\) … \({m}_{i,j}\) ], [ \({y}_{1}\) , \({y}_{2}\) … \({y}_{j}\) ]}. By utilizing this meta-feature dataset, a meta-learner can be trained. When a new multivariate time series dataset \({D}_{new}\) is introduced, the recommendation system can collect a meta-feature vector without any labels. Then, the system leverages its meta-learner capabilities to label this vector, resulting in a multi-labeled vector with each value representing the probability of the most appropriate deep-learning classifier for this new dataset. The final recommendation consists of the top deep-learning classifiers with the highest probability values. Efficiently constructing the meta-feature dataset M from raw data remains a significant challenge for the recommendation system. A previous study [ 2 ] tackled this issue by using a random sampling method to reduce the univariate time series dataset size. The reduced dataset was then used to compute the classification accuracy of each candidate classifier, and the outcome was considered a meta-attribute of the meta-feature dataset. Through the meta-feature dataset, a meta-learner can be trained. Meta-learning is a technique that involves learning how to learn and can be useful for algorithm selection or hyperparameter optimization. The idea behind meta-learning is to train a meta-learner with a small amount of data. However, when working with complex multivariate time series datasets, this can be a challenging task due to the dimensions of each instance. To simplify the process of creating the meta-feature dataset, I employed the random sampling method as in [ 2 ] and key channel selection methods described in [ 5 ]. The random sampling method reduces the number of instances in each dataset, while the key channel selection method reduces the number of dimensions in each instance. Figure 1 and Fig. 2 illustrate the scheme of the recommendation system I have developed. According to Fig. 1 , the raw data in each training dataset is initially padded with 0, ensuring that each time series has the same length. These padded datasets are then used to train candidate deep-learning classifiers, without any additional preprocessing. During the training process, the accuracy of each classifier for each padded dataset is calculated and recorded. After the training is complete, this computation result is used as the label for the meta-feature dataset. To reduce the size of the datasets, the padded datasets undergo random sampling and key channel selection. The candidate classifiers are then trained again using the reduced datasets, the accuracy of each classifier for each reduced dataset is used as the attribute of the meta-feature dataset. In the end, the meta-learner is trained using this labeled meta-feature dataset. When a new dataset is received, as depicted in Fig. 2 , it undergoes the identical processing as the training datasets. This process produces a meta-attribute vector that is devoid of any label. Once the processing is complete, the meta-learner illustrated in Fig. 1 will allocate multiple labels to this new meta-attribute vector. Each label signifies the likelihood of each classifier being an appropriate choice for the new dataset. The final output are top 3 deep-learning classifiers with maximum label value. 3. Experimentation The experiment consists of two distinct phases: the initial phase outlines the training process for a meta-learner, while the second phase evaluates the recommended outcome using a specific set of procedures. To validate the recommended result, two separate datasets are employed - one for training purposes and the other for validation. The ultimate goal is to test the overall effectiveness of the recommendation system and its ability to generalize to new datasets. 3.1. Training a Meta-learner 3.1.1 Training Datasets I conducted experiments on multivariate time series datasets from the UCR repository[ 4 ]. The repository comprises 30 different datasets that have varying characteristics, including time series lengths, number of classes, and application domains. For further information on these datasets, please refer to Table 1 . All datasets were acquired through the utilization of tsai, an open-source deep learning platform that employs Pytorch and fastai to deliver state-of-the-art methods for time series assignments like categorization and prediction[ 6 ]. After downloading, all datasets were standardized by applying uniform zero-padding and ensuring that each time series in every dataset had identical length. 3.1.2 Candidate Classifiers Twenty deep learning classifiers implemented in tsai had been selected as candidate classifiers. These classifiers include InceptionTime[ 7 ], FCN[ 8 ], LSTM_FCN, RNN_FCN, GRU_FCN, MLSTM_FCN, MGRU_FCN, TCN[ 9 ], XceptionTime[ 10 ], OmniScaleCNN[ 11 ], TST[ 12 ], TSiT[ 13 ], XCM[ 14 ], gMLP[ 15 ], TSSequencerPlus[ 16 ], PatchTST[ 17 ], ROCKET with RidgeClassifierCV[ 18 ], mWDN[ 19 ], MiniRocket[ 20 ], and ResNet[ 21 ]. 3.1.3 Experimental Set-up This training program, which utilizes 20 classifiers, is based on the tutorial notebook from tsai and is written in Python 3.11. Due to the requirement of running the program on 30 datasets to train the meta-learner, the execution time was reduced by running it on a server equipped with an RTX A100 GPU card boasting 40G of memory. First, the program was run on padded datasets, and the training results of about 20 classifiers were treated as the meta-label. This initial step took approximately 2 days. Next, I employed both random sampling and key channel selection techniques to decrease the size of each dataset. The subsample ratio was consistent with [ 2 ] and is displayed in Table 2 . For the key channel selection algorithm, I utilized Elbow Class Pairwise (ECP)[ 22 ], which is implemented in sktime, a Python library for time series analysis. The training outcome pertaining to the reduced training datasets served as the meta-attribute. This second phase of the process required approximately 8 hours to complete. Table 1 The details about 30 training datasets. Dataset Train Size Test Size Length No. of Classes No. of Dimensions Type ArticularyWordRecognition 275 300 144 25 9 MOTION AtrialFibrillation 15 15 640 3 2 ECG BasicMotions 40 40 100 4 6 HAR CharacterTrajectories 1422 1436 182 20 3 MOTION Cricket 108 72 1197 12 6 HAR DuckDuckGeese 60 40 270 5 1345 AUDIO EigenWorms 31 128 17984 5 6 MOTION Epilepsy 137 138 207 4 3 HAR Ering 30 270 65 6 4 HAR EthanolConcentration 261 263 1751 4 3 OTHER FaceDetection 5890 3524 62 2 144 EEG FingerMovements 316 100 50 2 28 EEG HandMovementDirection 160 74 400 4 10 EEG Handwriting 150 850 152 26 3 HAR Heartbeat 204 205 405 2 61 MOTION InsectWingbeat 30000 20000 30 10 200 AUDIO JapaneseVowels 270 370 29 9 12 AUDIO Libras 180 180 45 15 2 HAR LSST 2459 2466 36 14 6 OTHER MotorImagery 278 100 3000 2 64 EEG NATOPS 180 180 51 6 24 HAR PEMS-SF 267 173 144 7 963 MISC PenDigits 7494 3498 8 10 2 MOTION PhonemeSpectra 3315 3353 217 39 11 SOUND RacketSports 151 152 30 4 6 HAR SelfRegulationSCP1 268 293 896 2 6 EEG SelfRegulationSCP2 200 180 1152 2 7 EEG SpokenArabicDigits 6599 2199 93 10 13 SPEECH StandWalkJump 12 15 2500 3 4 ECG UWaveGestureLibrary 2238 2241 315 8 3 HAR Table 2 The subsample ratio[ 2 ] applied to a time series dataset with N instances Instances count Subsample ratio N < 100 0.80 100 ≤ N < 300 0.60 300 ≤ N < 800 0.40 800 ≤ N < 1500 0.20 1500 ≤ N < 5000 0.10 5000 ≤ N 0.05 During the initial training stages of certain classifiers, such as TST and TSiT, utilizing EigenWorms and MotorImagery datasets, an error concerning GPU memory overflow occurred. This was likely due to the lengthy time series involved. As a result, I removed these datasets and the resulting meta-feature dataset size is now 28*20, reflecting the number of training datasets and candidate classifiers, respectively. Finally, the meta-feature dataset was used to train a linear regression, and the resulting model was used as the meta-learner. The training program was implemented using scikit-learn, an open-source machine-learning library in Python, which was the same as [ 2 ]. 3.2. Recommendation Suitable Classifier 3.2.1 Validation Datasets In order to evaluate the meta-learner, I incorporated supplementary datasets obtained from Mustafa Baydogan. In this repository, 13 datasets are in "mat" format and are not accessible for download online. After reaching out to Baydogan via email, I was able to obtain this repository and convert the files to "npx" format using a specialized program. This enabled me to load them with tsai. To ensure consistency, I padded each dataset with zeros and standardized the length of each time series. For more detailed information on the 13 datasets, please see Table 3 . In order to validate the recommended results, 13 padded datasets were utilized to train 20 candidate classifiers and determine their classification accuracies. These results were established as the baseline for comparison. During the training process, it was discovered that the WalkvsRun dataset produced identical results in 18 of the 20 classifiers. In order to identify the top 3 most suitable classifiers for validating the recommended outcome, it was necessary to remove the WalkvsRun dataset from consideration, as it would not have provided an effective means of validation. Table 3 The details about 13 validation datasets. Dataset Train Size Test Size Length No. of Classes No. of Dimensions ArabicDigits 6600 2200 93 10 13 AUSLAN 1140 1425 136 95 22 CharacterTrajectories 300 2558 205 20 3 CMUsubject16 29 29 580 62 2 ECG 100 100 152 2 2 JapaneseVowels 270 370 29 9 12 KickvsPunch 16 10 841 2 62 Libras 180 180 45 15 2 NetFlow 803 534 997 2 4 PEMS 267 173 144 7 963 Uwave 200 4278 315 8 3 Wafer 298 896 198 2 6 WalkvsRun 28 16 1918 2 62 In the next step, the remaining validation datasets were utilized to create a meta-attribute, using both the random sampling method and the key channel selection method. However, during the selection of key channels, the Elbow Class Pairwise (ECP) algorithm encountered an error on five datasets - AUSLAN, Libras, PEMS, Uwave, and Wafer - due to the presence of a channel with all zeros. In the future, I will work on finding a solution to this issue. For now, I have removed these 5 datasets from the validation set. 3.2.2 Validating Recommended Result The validation datasets comprised of ArabicDigits, CharacterTrajectories, ECG, JapaneseVowels, KickvsPunch, NetFlow, and Wafer. As the meta-learner program used a random sampling approach, the suggested outcome varied every time it was executed. To ensure a comprehensive validation of the recommendation, I ran the program 10 times across these 7 datasets and ranked the classifiers based on the total number of occurrences. The following example shows how to run the meta-learner program 10 times in a couple lines of code: from recommend_classifier import recommend_classifier if __name__ == '__main__': recommend_result = recommend_classifier("must","ECG", "D:/project/studyproject/tsai/ESDL-MTSF/data/MUST-NPY/ECG",10) print("Top 3 result:" + str(recommend_result)) The inputs of the system are the repository name, the dataset name, the directory path saving the “npx” files, and the running times. The leading 3 classifiers are output and compared with the baseline result. For further information, please see Table 4 . Table 4 :Validating the recommended result Dataset Top 3 Recommended result Top 3 Baseline result ArabicDigits [InceptionTime] , FCN, [ResNet] [ResNet] , [InceptionTime] , mWDN CharacterTrajectories [InceptionTime] , ResNet, FCN ROCKET, [InceptionTime] , OmniScaleCNN ECG XCM, [InceptionTime] , TSSequencerPlus MiniRocket, [InceptionTime] , OmniScaleCNN JapaneseVowels [MiniRocket] , InceptionTime, FCN LSTM_FCN, MLSTM_FCN, [MiniRocket] Wafer FCN, [TSSequencerPlus] , XCM MiniRocket, mWDN, [TSSequencerPlus] NetFlow [MiniRocket] , PatchTST, TSSequencerPlus [MiniRocket] , ROCKET, ResNet KickvsPunch XceptionTime, InceptionTime, TST MiniRocket, XCM, TSSequencerPlus According to Table 4 , the recommendation system successfully identified the top 2 classifiers, InceptionTime and ResNet, for the ArabicDigits dataset, although their order was different from the baseline result. However, only the second-best classifier, InceptionTime, was identified for the CharacterTrajectories and ECG datasets. Similarly, MiniRocket and TSSequencerPlus were the third-best classifier for JapaneseVowels and Wafer, respectively. The system accurately identified MiniRocket as the most suitable classifier for the NetFlow dataset. Unfortunately, no suitable classifiers were found for the KickvsPunch dataset. 4. Conclusion In this paper, a recommendation system is presented that is designed to automatically identify the most suitable deep-learning classifier for multivariate time series classification problems. The main challenge of this system is the speedy construction of a meta-feature dataset. To address this, a random sampling method and key channel selection method were implemented to reduce the raw multivariate time series datasets and build the necessary meta-feature dataset. Using this meta-feature dataset, a meta-learner was trained which successfully recommended a suitable deep-learning classifier for some validation datasets, as demonstrated by experimental results. Nonetheless, there are still some unresolved issues that require further investigation, such as how to handle datasets with lengthy time series and channels with all zeros. These issues will be explored in future work. Declarations Author Contribution The paper is wrote by one author References Wolpert DH, Macready WG, Analysis F No free lunch theorems for search, Technical Report SFI-TR-95-02-010, The Santa Fe Institute,1996 Abanda A, Mori U, Lozano JA (2022) Time series classifier recommendation by a meta-learning approach, Pattern Recognition, vol. 128, Aug Chen Y, Keogh E, Hu B, Begum N, Bagnall A, Mueen A, Batista GE The UCR time series classification archive,2015. Bagnall A, Lines J, Vickers W, Keogh E The UEA and UCR Time Series Classification Repository, http://www.timeseriesclassification.com Dhariyal B, Nguyen TL, Ifrim G (2021) Fast Channel Selection for Scalable Multivariate Time Series Classification, In International Workshop on Advanced Analytics and Learning on Temporal Data. Springer, pp. 36–54 Ignacio Oguiza tsai - A state-of-the-art deep learning library for time series and sequential data, https://github.com/timeseriesAI/tsai Fawaz HI, Lucas B, Forestier G, Pelletier C, Schmidt DF, Weber J, Petitjean F InceptionTime: Finding AlexNet for Time Series Classification, arXiv preprint arXiv:1909.04939,2019. Wang Z, Yan W, Oates T (2017) Time series classification from scratch with deep neural networks: A strong baseline, In international joint conference on neural networks (IJCNN), pp. 1578–1585 Bai S, Kolter JZ, Koltun An empirical evaluation of generic convolutional and recurrent networks for sequence modeling, arXiv preprint arXiv:1803.01271,2018 Fawaz HI, Lucas B, Forestier G, Pelletier C, Schmidt DF, Weber J, Petitjean F InceptionTime: Finding AlexNet for Time Series Classification, arXiv preprint arXiv:1909.04939, 2019. Rußwurm M, Körner M (2019) Self-attention for raw optical satellite time series classification, arXiv preprint arXiv:1910.10536 George, Zerveas et al (2021) A Transformer-based Framework for Multivariate Time Series Representation Learning, In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’21), August 14–18 Alexey Dosovitskiy L, Beyer A, Kolesnikov D, Weissenborn X, Zhai T, Unterthiner M, Dehghani M, Minderer G, Heigold Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby, An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.11929,2020 Fauvel K, Lin T, Masson V, Fromont E, Termier A (2021) XCM: An Explainable Convolutional Neural Network for Multivariate Time Series Classification, Mathematics, vol. 9, Aug Liu H, Dai Z, So DR, Le QV (2021) Pay Attention to MLPs, arXiv preprint arXiv:2105.08050 Tatsunami Y, Taki M Sequencer: Deep LSTM for Image Classification, arXiv preprint arXiv:2205.01972,2022 Nie Y, Nguyen NH, Sinthong P, Kalagnanam J A Time Series is Worth 64 Words: Long-term Forecasting with Transformers, arXiv preprint arXiv:2211.14730,2022 Angus Dempster François, Petitjean GI, Webb ROCKET Exceptionally fast and accurate time series classification using random convolutional kernels, arXiv:1910.13051,2019 Wang J, Wang Z, Li J, Wu J Multilevel wavelet decomposition network for interpretable time series analysis, In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 2437–2446, July, 2018 Tan CW, Dempster A, Bergmeir C, Webb GI MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification, arXiv:2102.00457,2021 Zhiguang Wang W, Oates YT Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline, arXiv:1611.06455,2016 Bhaskar, Dhariyal et al (2021) Fast Channel Selection for Scalable Multivariate Time Series Classification. AALTD, ECML-PKDD, Springer, Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3999075","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":275564221,"identity":"f7ebf1d5-51cc-480a-aa5f-582d6e7ab7aa","order_by":0,"name":"Rui Gan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYBACxvbmw4//GNjw2Lc3EKmFuedYmgFPRZqcAc8BIrWwz/AxkOA5c9jYQCKBSC28M3gMDCTbmBO3Sz7eeIOhxiaaoBbJ2W0FDwzb2BJ3zk4rtmA4lpbbQEiL4ZzDGwwS23gSG27nmEkwNhwmrMX+RoKBxME2icSGm2eI1MI4I8VAsuGMgbHBDR5itQAD2ZihIkFOsgfolwRi/AKOSgaD/zz87Ic33vhQY0NYCzIgPmqQtJCqYxSMglEwCkYGAADC6UMmqd4ZLwAAAABJRU5ErkJggg==","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Rui","middleName":"","lastName":"Gan","suffix":""}],"badges":[],"createdAt":"2024-02-29 08:46:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3999075/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3999075/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":52101046,"identity":"cda1ea76-57c3-4a50-898b-66e69dc5d5a3","added_by":"auto","created_at":"2024-03-06 19:00:37","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1291838,"visible":true,"origin":"","legend":"\u003cp\u003eThe scheme for training meta-learner\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3999075/v1/2d67af30fbc13787c89161be.jpeg"},{"id":52101048,"identity":"644dc6f2-c406-4e04-8f1c-f7f4327244cb","added_by":"auto","created_at":"2024-03-06 19:00:39","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1007036,"visible":true,"origin":"","legend":"\u003cp\u003eThe scheme about recommending classifier\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3999075/v1/3f33bc0a1f930b53287e81ed.jpeg"},{"id":52686529,"identity":"57108ec0-276d-4ede-8b56-ae9ef32c3086","added_by":"auto","created_at":"2024-03-14 13:50:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":591286,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3999075/v1/26a0a020-667e-413b-bcc4-98de2ed54922.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Suitable Deep Learning Classifier Recommendation for Multi-variate Time Series Classification","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eTime series classification is a highly versatile tool that has been applied in various fields, including bioinformatics and finance. Since each domain has its own distinct time series data format, the classification problem can be categorized as univariate time series classification or multivariate time series classification. As the \"No Free Lunch\" theorem[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] suggests, there's no one algorithm that can provide the best solution for all problems. Therefore, discovering a suitable classifier for a particular time series classification problem necessitates a trial-and-error approach. Is there a method to streamline this process and make finding a suitable classifier simpler?\u003c/p\u003e \u003cp\u003eYes, that is correct. A study by Abanda.A, published in [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], a time series classifier recommendation (TSCR) system was developed using a meta-learning approach to identify the suitable classifier for univariate time series datasets. One of the primary challenges for the TSCR system was generating a meta-feature dataset, known as landmarker, in a timely fashion. Once this landmarker was acquired, a meta-learner could be trained. The author of [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] utilized 112 subsample univariate time datasets from the UCR repository [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] to train 24 classifiers and determine the classification accuracy of each one. These calculations were treated as attributes of each instance in the meta-feature dataset, resulting in a meta-feature dataset size of 112*24, with 112 denoting the number of time series datasets and 24 denoting the number of classifiers. The outcomes of extensive experimentation on 24 classifier algorithms publicly available in [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] were then utilized to label each instance in the meta-feature dataset. Therefore, the issue of classifier recommendation can be seen as a multi-label learning problem, with each label indicating the likelihood of a classifier being well-suited for a specific dataset.\u003c/p\u003e \u003cp\u003eAt present, the TSCR system is the only thoroughly researched option for suggesting time series classifiers. However, it is limited to univariate time series. Furthermore, out of the 24 candidate classifiers in [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], only 2 are deep-learning classifiers. Multivariate time series datasets are inherently more complex due to their multiple dimensions, making them more intricate than univariate datasets. Additionally, recent research has introduced new deep-learning algorithms for solving time series classification problems. In this article, I will introduce a multivariate time series classification recommendation (MTSCR) system called DL-MTSCR with a meta-learning approach. This system can automatically identify the suitable deep-learning classifier for multivariate time series classification. For those interested, the source code can be downloaded from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gitee.com/kengan1013/esdl-mtsf\u003c/span\u003e\u003cspan address=\"https://gitee.com/kengan1013/esdl-mtsf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe DL-MTSCR system faces a crucial hurdle in constructing a meta-feature dataset for complex data. To overcome this challenge, the system combines two efficient methods that simplify multivariate time series data and facilitate the creation of a meta-feature dataset. The rest of the paper is organized as follows: Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the framework for the recommendation system while highlighting the core idea of rapidly constructing a meta-feature dataset for multivariate time series data. Section \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003e3\u003c/span\u003e showcases the results of experimentation, and finally, Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e4\u003c/span\u003e provides concluding remarks and suggests possible directions for future research.\u003c/p\u003e"},{"header":"2. Deep-Learning Classifier Recommendation in DL-MTSCR","content":"\u003cp\u003eThe difference between univariate time series and multivariate time series lies in their respective number of dimensions. In an univariate time series dataset, each instance has a dimension of 1, while a multivariate time series dataset has a dimension greater than 1. As such, a recommendation system for multivariate time series classifiers can be developed in the following manner:\u003c/p\u003e \u003cp\u003eTo begin, giving a set of multivariate time series datasets \u003cem\u003eD\u003c/em\u003e= {\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{1}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{i}\\)\u003c/span\u003e\u003c/span\u003e}. Each dataset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{i}\\)\u003c/span\u003e\u003c/span\u003e contains a set of \u003cem\u003en\u003c/em\u003e instance label pairs, represented by \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{i}\\)\u003c/span\u003e\u003c/span\u003e = {(\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{1}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{1}\\)\u003c/span\u003e\u003c/span\u003e), (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{2}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{2}\\)\u003c/span\u003e\u003c/span\u003e). .. (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{n}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{n}\\)\u003c/span\u003e\u003c/span\u003e)}, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{n}\\)\u003c/span\u003e\u003c/span\u003e is the label associated to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{n}\\)\u003c/span\u003e\u003c/span\u003e. Each \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{n}\\)\u003c/span\u003e\u003c/span\u003e is a list of \u003cem\u003ed\u003c/em\u003e vectors, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{n}\\)\u003c/span\u003e\u003c/span\u003e= [\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{2}\\)\u003c/span\u003e\u003c/span\u003e \u0026middot; \u0026middot; \u0026middot;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{d}\\)\u003c/span\u003e\u003c/span\u003e]. Each vector \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{d}\\)\u003c/span\u003e\u003c/span\u003e is a time series data vector with \u003cem\u003em\u003c/em\u003e feature value, represented by \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{d}\\)\u003c/span\u003e\u003c/span\u003e= [\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{m}\\)\u003c/span\u003e\u003c/span\u003e]. Therefore, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{n}\\)\u003c/span\u003e\u003c/span\u003e can be viewed as a matrix with \u003cem\u003ed\u003c/em\u003e rows and \u003cem\u003em\u003c/em\u003e columns, where \u003cem\u003ed\u003c/em\u003e is the number of time series and \u003cem\u003em\u003c/em\u003e is the length of each time series. Additionally, there is a set of deep-learning classifiers C= {\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{j}\\)\u003c/span\u003e\u003c/span\u003e}.\u003c/p\u003e \u003cp\u003eNext, the recommendation system utilizes all multivariate time series datasets in \u003cem\u003eD\u003c/em\u003e to construct a meta-feature dataset \u003cem\u003eM\u003c/em\u003e= {(\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{1}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Y}_{1}\\)\u003c/span\u003e\u003c/span\u003e), (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{2}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Y}_{2}\\)\u003c/span\u003e\u003c/span\u003e)\u0026hellip;(\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{i}\\)\u003c/span\u003e\u003c/span\u003e,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Y}_{i}\\)\u003c/span\u003e\u003c/span\u003e)}, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{i}\\)\u003c/span\u003e\u003c/span\u003e is a feature vector pertaining to dataset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{i}\\)\u003c/span\u003e\u003c/span\u003e. Therefore, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{i}\\)\u003c/span\u003e\u003c/span\u003e= [\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e] where each \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e is a meta-attribute value about dataset\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({ D}_{i}\\)\u003c/span\u003e\u003c/span\u003eand deep-learning classifier \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{j}\\)\u003c/span\u003e\u003c/span\u003e. Additionally, each \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{i}\\)\u003c/span\u003e\u003c/span\u003e also has a label vector \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Y}_{i}\\)\u003c/span\u003e\u003c/span\u003e, each feature value in \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({Y}_{i}\\)\u003c/span\u003e\u003c/span\u003e corresponds to the classification accuracy of each classifier \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{j}\\)\u003c/span\u003e\u003c/span\u003e on dataset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{i}\\)\u003c/span\u003e\u003c/span\u003e. Consequently, the labeled meta-feature dataset can be expressed as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({M}_{i}\\)\u003c/span\u003e\u003c/span\u003e= {[\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({m}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e], [\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{2}\\)\u003c/span\u003e\u003c/span\u003e\u0026hellip;\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{j}\\)\u003c/span\u003e\u003c/span\u003e]}. By utilizing this meta-feature dataset, a meta-learner can be trained.\u003c/p\u003e \u003cp\u003eWhen a new multivariate time series dataset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({D}_{new}\\)\u003c/span\u003e\u003c/span\u003eis introduced, the recommendation system can collect a meta-feature vector without any labels. Then, the system leverages its meta-learner capabilities to label this vector, resulting in a multi-labeled vector with each value representing the probability of the most appropriate deep-learning classifier for this new dataset. The final recommendation consists of the top deep-learning classifiers with the highest probability values.\u003c/p\u003e \u003cp\u003eEfficiently constructing the meta-feature dataset \u003cem\u003eM\u003c/em\u003e from raw data remains a significant challenge for the recommendation system. A previous study [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] tackled this issue by using a random sampling method to reduce the univariate time series dataset size. The reduced dataset was then used to compute the classification accuracy of each candidate classifier, and the outcome was considered a meta-attribute of the meta-feature dataset. Through the meta-feature dataset, a meta-learner can be trained.\u003c/p\u003e \u003cp\u003eMeta-learning is a technique that involves learning how to learn and can be useful for algorithm selection or hyperparameter optimization. The idea behind meta-learning is to train a meta-learner with a small amount of data. However, when working with complex multivariate time series datasets, this can be a challenging task due to the dimensions of each instance. To simplify the process of creating the meta-feature dataset, I employed the random sampling method as in [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] and key channel selection methods described in [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The random sampling method reduces the number of instances in each dataset, while the key channel selection method reduces the number of dimensions in each instance. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrate the scheme of the recommendation system I have developed.\u003c/p\u003e \u003cp\u003eAccording to Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the raw data in each training dataset is initially padded with 0, ensuring that each time series has the same length. These padded datasets are then used to train candidate deep-learning classifiers, without any additional preprocessing. During the training process, the accuracy of each classifier for each padded dataset is calculated and recorded. After the training is complete, this computation result is used as the label for the meta-feature dataset. To reduce the size of the datasets, the padded datasets undergo random sampling and key channel selection. The candidate classifiers are then trained again using the reduced datasets, the accuracy of each classifier for each reduced dataset is used as the attribute of the meta-feature dataset. In the end, the meta-learner is trained using this labeled meta-feature dataset.\u003c/p\u003e \u003cp\u003eWhen a new dataset is received, as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, it undergoes the identical processing as the training datasets. This process produces a meta-attribute vector that is devoid of any label. Once the processing is complete, the meta-learner illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e will allocate multiple labels to this new meta-attribute vector. Each label signifies the likelihood of each classifier being an appropriate choice for the new dataset. The final output are top 3 deep-learning classifiers with maximum label value.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"3. Experimentation","content":"\u003cp\u003eThe experiment consists of two distinct phases: the initial phase outlines the training process for a meta-learner, while the second phase evaluates the recommended outcome using a specific set of procedures. To validate the recommended result, two separate datasets are employed - one for training purposes and the other for validation. The ultimate goal is to test the overall effectiveness of the recommendation system and its ability to generalize to new datasets.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Training a Meta-learner\u003c/h2\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e3.1.1 Training Datasets\u003c/h2\u003e \u003cp\u003eI conducted experiments on multivariate time series datasets from the UCR repository[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The repository comprises 30 different datasets that have varying characteristics, including time series lengths, number of classes, and application domains. For further information on these datasets, please refer to Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eAll datasets were acquired through the utilization of tsai, an open-source deep learning platform that employs Pytorch and fastai to deliver state-of-the-art methods for time series assignments like categorization and prediction[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. After downloading, all datasets were standardized by applying uniform zero-padding and ensuring that each time series in every dataset had identical length.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e3.1.2 Candidate Classifiers\u003c/h2\u003e \u003cp\u003eTwenty deep learning classifiers implemented in tsai had been selected as candidate classifiers. These classifiers include InceptionTime[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], FCN[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], LSTM_FCN, RNN_FCN, GRU_FCN, MLSTM_FCN, MGRU_FCN, TCN[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], XceptionTime[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], OmniScaleCNN[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], TST[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], TSiT[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], XCM[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], gMLP[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], TSSequencerPlus[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], PatchTST[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], ROCKET with RidgeClassifierCV[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], mWDN[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], MiniRocket[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], and ResNet[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.1.3 Experimental Set-up\u003c/h2\u003e \u003cp\u003eThis training program, which utilizes 20 classifiers, is based on the tutorial notebook from tsai and is written in Python 3.11. Due to the requirement of running the program on 30 datasets to train the meta-learner, the execution time was reduced by running it on a server equipped with an RTX A100 GPU card boasting 40G of memory.\u003c/p\u003e \u003cp\u003eFirst, the program was run on padded datasets, and the training results of about 20 classifiers were treated as the meta-label. This initial step took approximately 2 days.\u003c/p\u003e \u003cp\u003eNext, I employed both random sampling and key channel selection techniques to decrease the size of each dataset. The subsample ratio was consistent with [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] and is displayed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. For the key channel selection algorithm, I utilized Elbow Class Pairwise (ECP)[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], which is implemented in sktime, a Python library for time series analysis. The training outcome pertaining to the reduced training datasets served as the meta-attribute. This second phase of the process required approximately 8 hours to complete.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe details about 30 training datasets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrain Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTest Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLength\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNo. of Classes\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo. of Dimensions\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eType\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArticularyWordRecognition\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e275\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMOTION\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAtrialFibrillation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e640\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eECG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasicMotions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacterTrajectories\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1422\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1436\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMOTION\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCricket\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e108\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1197\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuckDuckGeese\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1345\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAUDIO\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEigenWorms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e128\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e17984\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMOTION\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEpilepsy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e138\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e207\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEthanolConcentration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e261\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e263\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1751\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eOTHER\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFaceDetection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5890\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3524\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFingerMovements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e316\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHandMovementDirection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHandwriting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e150\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeartbeat\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e204\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e205\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e405\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMOTION\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInsectWingbeat\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e20000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAUDIO\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapaneseVowels\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e370\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAUDIO\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLibras\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLSST\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2459\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2466\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eOTHER\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMotorImagery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNATOPS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePEMS-SF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e267\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e173\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMISC\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePenDigits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7494\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3498\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMOTION\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhonemeSpectra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3315\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3353\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e217\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSOUND\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRacketSports\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e151\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSelfRegulationSCP1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e293\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e896\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSelfRegulationSCP2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eEEG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpokenArabicDigits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6599\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2199\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSPEECH\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStandWalkJump\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eECG\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUWaveGestureLibrary\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2238\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2241\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e315\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHAR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe subsample ratio[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] applied to a time series dataset with N instances\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInstances count\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSubsample ratio\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u0026thinsp;\u0026lt;\u0026thinsp;100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e100\u0026thinsp;\u0026le;\u0026thinsp;N\u0026thinsp;\u0026lt;\u0026thinsp;300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e300\u0026thinsp;\u0026le;\u0026thinsp;N\u0026thinsp;\u0026lt;\u0026thinsp;800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e800\u0026thinsp;\u0026le;\u0026thinsp;N\u0026thinsp;\u0026lt;\u0026thinsp;1500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1500\u0026thinsp;\u0026le;\u0026thinsp;N\u0026thinsp;\u0026lt;\u0026thinsp;5000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5000\u0026thinsp;\u0026le;\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eDuring the initial training stages of certain classifiers, such as TST and TSiT, utilizing EigenWorms and MotorImagery datasets, an error concerning GPU memory overflow occurred. This was likely due to the lengthy time series involved. As a result, I removed these datasets and the resulting meta-feature dataset size is now 28*20, reflecting the number of training datasets and candidate classifiers, respectively.\u003c/p\u003e \u003cp\u003eFinally, the meta-feature dataset was used to train a linear regression, and the resulting model was used as the meta-learner. The training program was implemented using scikit-learn, an open-source machine-learning library in Python, which was the same as [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Recommendation Suitable Classifier\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Validation Datasets\u003c/h2\u003e \u003cp\u003eIn order to evaluate the meta-learner, I incorporated supplementary datasets obtained from Mustafa Baydogan. In this repository, 13 datasets are in \"mat\" format and are not accessible for download online. After reaching out to Baydogan via email, I was able to obtain this repository and convert the files to \"npx\" format using a specialized program. This enabled me to load them with tsai. To ensure consistency, I padded each dataset with zeros and standardized the length of each time series. For more detailed information on the 13 datasets, please see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eIn order to validate the recommended results, 13 padded datasets were utilized to train 20 candidate classifiers and determine their classification accuracies. These results were established as the baseline for comparison. During the training process, it was discovered that the WalkvsRun dataset produced identical results in 18 of the 20 classifiers. In order to identify the top 3 most suitable classifiers for validating the recommended outcome, it was necessary to remove the WalkvsRun dataset from consideration, as it would not have provided an effective means of validation.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe details about 13 validation datasets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrain Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTest Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLength\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNo. of Classes\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo. of Dimensions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArabicDigits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAUSLAN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1140\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1425\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e136\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacterTrajectories\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2558\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e205\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCMUsubject16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e580\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e152\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapaneseVowels\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e370\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKickvsPunch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e841\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e62\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLibras\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNetFlow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e534\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e997\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePEMS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e267\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e173\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e963\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUwave\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e315\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWafer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e298\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e896\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e198\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWalkvsRun\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e62\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the next step, the remaining validation datasets were utilized to create a meta-attribute, using both the random sampling method and the key channel selection method. However, during the selection of key channels, the Elbow Class Pairwise (ECP) algorithm encountered an error on five datasets - AUSLAN, Libras, PEMS, Uwave, and Wafer - due to the presence of a channel with all zeros. In the future, I will work on finding a solution to this issue. For now, I have removed these 5 datasets from the validation set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Validating Recommended Result\u003c/h2\u003e \u003cp\u003eThe validation datasets comprised of ArabicDigits, CharacterTrajectories, ECG, JapaneseVowels, KickvsPunch, NetFlow, and Wafer. As the meta-learner program used a random sampling approach, the suggested outcome varied every time it was executed. To ensure a comprehensive validation of the recommendation, I ran the program 10 times across these 7 datasets and ranked the classifiers based on the total number of occurrences. The following example shows how to run the meta-learner program 10 times in a couple lines of code:\u003c/p\u003e \u003cp\u003efrom recommend_classifier import recommend_classifier\u003c/p\u003e \u003cp\u003eif __name__ == '__main__':\u003c/p\u003e \u003cp\u003erecommend_result\u0026thinsp;=\u0026thinsp;recommend_classifier(\"must\",\"ECG\", \"D:/project/studyproject/tsai/ESDL-MTSF/data/MUST-NPY/ECG\",10)\u003c/p\u003e \u003cp\u003eprint(\"Top 3 result:\" + str(recommend_result))\u003c/p\u003e \u003cp\u003eThe inputs of the system are the repository name, the dataset name, the directory path saving the \u0026ldquo;npx\u0026rdquo; files, and the running times. The leading 3 classifiers are output and compared with the baseline result. For further information, please see Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e:Validating the recommended result\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTop 3 Recommended result\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTop 3 Baseline result\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArabicDigits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e[InceptionTime]\u003c/b\u003e,\u003c/p\u003e \u003cp\u003eFCN, \u003cb\u003e[ResNet]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e[ResNet]\u003c/b\u003e, \u003cb\u003e[InceptionTime]\u003c/b\u003e, mWDN\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacterTrajectories\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e[InceptionTime]\u003c/b\u003e, ResNet, FCN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eROCKET, \u003cb\u003e[InceptionTime]\u003c/b\u003e, OmniScaleCNN\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eXCM, \u003cb\u003e[InceptionTime]\u003c/b\u003e,\u003c/p\u003e \u003cp\u003eTSSequencerPlus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMiniRocket, \u003cb\u003e[InceptionTime]\u003c/b\u003e, OmniScaleCNN\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapaneseVowels\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e[MiniRocket]\u003c/b\u003e, InceptionTime, FCN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLSTM_FCN, MLSTM_FCN, \u003cb\u003e[MiniRocket]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWafer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFCN, \u003cb\u003e[TSSequencerPlus]\u003c/b\u003e, XCM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMiniRocket, mWDN, \u003cb\u003e[TSSequencerPlus]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNetFlow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e[MiniRocket]\u003c/b\u003e, PatchTST, TSSequencerPlus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e[MiniRocket]\u003c/b\u003e, ROCKET, ResNet\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKickvsPunch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eXceptionTime, InceptionTime, TST\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMiniRocket, XCM, TSSequencerPlus\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAccording to Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the recommendation system successfully identified the top 2 classifiers, InceptionTime and ResNet, for the ArabicDigits dataset, although their order was different from the baseline result. However, only the second-best classifier, InceptionTime, was identified for the CharacterTrajectories and ECG datasets. Similarly, MiniRocket and TSSequencerPlus were the third-best classifier for JapaneseVowels and Wafer, respectively. The system accurately identified MiniRocket as the most suitable classifier for the NetFlow dataset. Unfortunately, no suitable classifiers were found for the KickvsPunch dataset.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Conclusion","content":"\u003cp\u003eIn this paper, a recommendation system is presented that is designed to automatically identify the most suitable deep-learning classifier for multivariate time series classification problems. The main challenge of this system is the speedy construction of a meta-feature dataset. To address this, a random sampling method and key channel selection method were implemented to reduce the raw multivariate time series datasets and build the necessary meta-feature dataset. Using this meta-feature dataset, a meta-learner was trained which successfully recommended a suitable deep-learning classifier for some validation datasets, as demonstrated by experimental results. Nonetheless, there are still some unresolved issues that require further investigation, such as how to handle datasets with lengthy time series and channels with all zeros. These issues will be explored in future work.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eThe paper is wrote by one author\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWolpert DH, Macready WG, Analysis F No free lunch theorems for search, Technical Report SFI-TR-95-02-010, The Santa Fe Institute,1996\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbanda A, Mori U, Lozano JA (2022) Time series classifier recommendation by a meta-learning approach, Pattern Recognition, vol. 128, Aug\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Keogh E, Hu B, Begum N, Bagnall A, Mueen A, Batista GE The UCR time series classification archive,2015.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBagnall A, Lines J, Vickers W, Keogh E The UEA and UCR Time Series Classification Repository, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.timeseriesclassification.com\u003c/span\u003e\u003cspan address=\"http://www.timeseriesclassification.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDhariyal B, Nguyen TL, Ifrim G (2021) Fast Channel Selection for Scalable Multivariate Time Series Classification, In International Workshop on Advanced Analytics and Learning on Temporal Data. Springer, pp. 36\u0026ndash;54\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIgnacio Oguiza tsai - A state-of-the-art deep learning library for time series and sequential data, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/timeseriesAI/tsai\u003c/span\u003e\u003cspan address=\"https://github.com/timeseriesAI/tsai\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFawaz HI, Lucas B, Forestier G, Pelletier C, Schmidt DF, Weber J, Petitjean F InceptionTime: Finding AlexNet for Time Series Classification, arXiv preprint arXiv:1909.04939,2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Yan W, Oates T (2017) Time series classification from scratch with deep neural networks: A strong baseline, In international joint conference on neural networks (IJCNN), pp. 1578\u0026ndash;1585\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai S, Kolter JZ, Koltun An empirical evaluation of generic convolutional and recurrent networks for sequence modeling, arXiv preprint arXiv:1803.01271,2018\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFawaz HI, Lucas B, Forestier G, Pelletier C, Schmidt DF, Weber J, Petitjean F InceptionTime: Finding AlexNet for Time Series Classification, arXiv preprint arXiv:1909.04939, 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRu\u0026szlig;wurm M, K\u0026ouml;rner M (2019) Self-attention for raw optical satellite time series classification, arXiv preprint arXiv:1910.10536\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGeorge, Zerveas et al (2021) A Transformer-based Framework for Multivariate Time Series Representation Learning, In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD \u0026rsquo;21), August 14\u0026ndash;18\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlexey Dosovitskiy L, Beyer A, Kolesnikov D, Weissenborn X, Zhai T, Unterthiner M, Dehghani M, Minderer G, Heigold Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby, An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.11929,2020\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFauvel K, Lin T, Masson V, Fromont E, Termier A (2021) XCM: An Explainable Convolutional Neural Network for Multivariate Time Series Classification, Mathematics, vol. 9, Aug\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu H, Dai Z, So DR, Le QV (2021) Pay Attention to MLPs, arXiv preprint arXiv:2105.08050\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTatsunami Y, Taki M Sequencer: Deep LSTM for Image Classification, arXiv preprint arXiv:2205.01972,2022\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNie Y, Nguyen NH, Sinthong P, Kalagnanam J A Time Series is Worth 64 Words: Long-term Forecasting with Transformers, arXiv preprint arXiv:2211.14730,2022\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAngus Dempster Fran\u0026ccedil;ois, Petitjean GI, Webb ROCKET Exceptionally fast and accurate time series classification using random convolutional kernels, arXiv:1910.13051,2019\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J, Wang Z, Li J, Wu J Multilevel wavelet decomposition network for interpretable time series analysis, In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery \u0026amp; Data Mining, pp. 2437\u0026ndash;2446, July, 2018\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan CW, Dempster A, Bergmeir C, Webb GI MultiRocket: Multiple pooling operators and transformations for fast and effective time series classification, arXiv:2102.00457,2021\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhiguang Wang W, Oates YT Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline, arXiv:1611.06455,2016\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhaskar, Dhariyal et al (2021) Fast Channel Selection for Scalable Multivariate Time Series Classification. AALTD, ECML-PKDD, Springer,\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Multivariate Time Series Classification, Metal-Learning, Deep-Learning","lastPublishedDoi":"10.21203/rs.3.rs-3999075/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3999075/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTime series classification can be categorized into two main types: univariate and multivariate. The major difference between the two is that multivariate datasets have multiple dimensions, which makes them more intricate than univariate datasets. Selecting a suitable classifier for multivariate time series classification can be challenging, as it involves trying out several algorithms to handle the intricate data. While some studies have been conducted on automatically finding suitable classifiers for univariate datasets, there are currently no such studies for multivariate ones. To fill this gap, this paper introduces a recommendation system for deep-learning classifiers that are suitable for multivariate time series classification. The experiments demonstrate that this recommendation system can automatically identify the suitable deep-learning classifier for some multivariate time series datasets.\u003c/p\u003e","manuscriptTitle":"Suitable Deep Learning Classifier Recommendation for Multi-variate Time Series Classification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-06 19:00:32","doi":"10.21203/rs.3.rs-3999075/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2183d6c8-00a3-4a32-aaf4-3a7d7a92360c","owner":[],"postedDate":"March 6th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-03-14T13:42:25+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-06 19:00:32","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3999075","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3999075","identity":"rs-3999075","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}