Dataset for onboard road-type classification using Machine-Learning Clustering models in a Renault Clio III vehicle

Data repository with the raw and synthesized data of a journal paper under review. The dataset provided is obtained from a Renault Clio III vehicle (model year 2007) by accessing the CAN bus through the OBD‑II port. Table 1 lists all the parameters extracted from the CAN bus, comprising approximately 40,000 data records collected during driving sessions that covered different types of roads, including urban roads, conventional roads, and highways, along a route from the city of Malaga to Cártama in southern Spain, spanning a total distance of 56.8 km. The weather conditions during the data acquisition period featured temperatures ranging between 22 °C and 25 °C and no precipitation in the spring of 2025. Table 1. Extracted signals from the CAN bus of a Renault Clio III vehicle included in the dataset. Description Parameter Value Unit Steering wheel angle SWA 27,545 (left) – 38,116 (right) deg Steering wheel angular velocity SWAV 0 – 32,767 deg/s Revolutions per minute RPM 0 – 8,192 rpm Throttle position Accel 16 (idle) - 252 (full pressed) - Vehicle speed Speed 0 – 655 km/h Travelled distance since ignition Distance 0 – 65,535 m Engine temperature EngTemp 0 – 255 ºC Since the data processing is carried out using clustering algorithms within the framework of Machine Learning (ML) techniques, computed values are incorporated into the dataset to describe the micro‑clusters formed from all points stored in the point buffer retrieved from the CAN bus, together with the macro‑clusters subsequently generated to update the centroid using the information contained in the micro‑cluster buffer. However, the data provided can also be used with other processing techniques in alternative lines of research. The dataset includes the following data files, which contain the parameters listed in Table 1 and incorporate the parameter Time as a sample index to indicate the arrival order of the data samples. Training data is composed of 12 comma-separated values (CSV) files with data collected during the driving sessions, as described in Table 2. Each file includes records for nine parameters that describe their state at each time instant, and whose identifiers are provided in the first row, such as Time, SWA, SWAV, RPM, Accel, Speed, Distance, and EngTemp. Table 2. Training data files used in the study. Name of the data file Number of data records Range of the parameter Time training_s1.csv 706 11 – 363.5 training_s2.csv 1,168 5.5 – 589.99 training_s3.csv 6,137 631.5 – 3,699.5 training_s4.csv 8,668 6.5 – 4,340 training_s5.csv 1,423 103.5 – 814.5 training_s6.csv 2,210 503.5 – 1,608 training_s7.csv 181 11 - 101 training_s8.csv 413 10 – 216 training_s9.csv 5,306 5 – 2657.5 training_s10.csv 1,402 8.5 – 709 training_s11.csv 3,650 13.5 – 1,923 training_s12.csv 5,505 1,679.5 – 4,431.5 Testing data is composed of 11 CSV files with the results of the micro-clustering and macro-clustering processes obtained during the driving sessions using different damping coefficients (DC) of between 36 and 48, as described in Table 3, where the identifiers of the parameters are provided in the first row of each file. The micro-clustering records include the parameter Time, the computed values of each CAN-bus signal for each micro-cluster (Speed, Accel, and SWAV), the assigned macro-cluster (AssignedCluster), and the label applied to the micro-cluster by the experimenter (Label). The same files also contain the macro-clustering results, which consist of the computed centroid position for each macro-cluster (Speed_MC, Accel_MC, and SWAV_MC) and their corresponding identifier (MacroCluster). Table 3. Testing data files used in the study. Name of the data file Number of data records Range of the parameter Time testing_dc_36_s1.csv 454 1 – 426 testing_dc_36_s2.csv 562 2 – 93 testing_dc_36_s3.csv 272 2 – 169 testing_dc_36_s4.csv 219 2 – 136 testing_dc_36_s5.csv 414 4 – 253 testing_dc_48_s1.csv 625 4 – 380 testing_dc_48_s2.csv 278 27 – 196 testing_dc_48_s3.csv 178 12 – 117 testing_dc_48_s4.csv 211 5 – 131 testing_dc_48_s5.csv 126 4 – 81 testing_dc_48_s6.csv 132 2 – 81 “Init.txt” is a configuration file that contains the initial values of the macro-clusters prior to each driving session. Each macro-cluster is initialized with the following values: “C=0” for the identifier, “S=00023” for speed, “A=033” for acceleration, and “SWAV=00359” for steering wheel rotation velocity.本数据集存储了某篇待投稿期刊论文的原始与合成数据。 本次采集的数据集来自一辆雷诺Clio III（2007款）车辆，通过OBD-II接口接入其CAN总线获取数据。表1列出了从该CAN总线提取的全部参数，本次共采集约4万条行驶工况数据，覆盖城市道路、普通公路与高速公路，行驶路线为西班牙南部马拉加至卡尔塔马，总里程56.8千米。数据采集于2025年春季，环境气温介于22℃至25℃之间，无降水。 表1 本数据集包含的雷诺Clio III车辆CAN总线提取信号 | 描述 | 参数 | 数值范围 | 单位 | | --- | --- | --- | --- | | 方向盘转角 | SWA | 27545（左）– 38116（右） | 度 | | 方向盘角速度 | SWAV | 0 – 32767 | 度/秒 | | 发动机转速 | RPM | 0 – 8192 | 转/分钟 | | 节气门位置 | Accel | 16（怠速）– 252（全开） | - | | 车辆速度 | Speed | 0 – 655 | 千米/小时 | | 点火后累计行驶里程 | Distance | 0 – 65535 | 米 | | 发动机温度 | EngTemp | 0 – 255 | 摄氏度 | 本数据集的处理基于机器学习（Machine Learning, ML）框架下的聚类算法，其中包含了从CAN总线获取的点缓存中所有数据点生成的微簇相关计算值，以及后续基于微簇缓存信息更新质心得到的宏簇相关计算值。此外，本数据集也可应用于其他研究方向的不同处理技术。 本数据集包含以下数据文件，均包含表1所列参数，并新增Time参数作为样本索引，用于标识数据样本的到达顺序。 ### 训练数据 训练数据由12个逗号分隔值（Comma-Separated Values, CSV）文件组成，包含前文所述行驶工况下采集的数据，详情见表2。每个文件包含9个描述车辆瞬时状态的参数，文件首行标注参数标识符，包括Time、SWA、SWAV、RPM、Accel、Speed、Distance与EngTemp。 表2 本研究使用的训练数据文件 | 数据文件名 | 数据记录数 | Time参数范围 | | --- | --- | --- | | training_s1.csv | 706 | 11 – 363.5 | | training_s2.csv | 1168 | 5.5 – 589.99 | | training_s3.csv | 6137 | 631.5 – 3699.5 | | training_s4.csv | 8668 | 6.5 – 4340 | | training_s5.csv | 1423 | 103.5 – 814.5 | | training_s6.csv | 2210 | 503.5 – 1608 | | training_s7.csv | 181 | 11 – 101 | | training_s8.csv | 413 | 10 – 216 | | training_s9.csv | 5306 | 5 – 2657.5 | | training_s10.csv | 1402 | 8.5 – 709 | | training_s11.csv | 3650 | 13.5 – 1923 | | training_s12.csv | 5505 | 1679.5 – 4431.5 | ### 测试数据 测试数据由11个CSV文件组成，包含采用36至48之间不同阻尼系数（Damping Coefficient, DC）进行行驶工况聚类得到的微簇与宏簇处理结果，详情见表3，每个文件首行标注参数标识符。微簇记录包含Time参数、各微簇对应的CAN总线信号计算值（Speed、Accel与SWAV）、分配的宏簇编号（AssignedCluster）以及实验人员为微簇赋予的标签（Label）。同时，文件中还包含宏簇处理结果，即各宏簇的计算质心位置（Speed_MC、Accel_MC与SWAV_MC）及其对应标识符（MacroCluster）。 表3 本研究使用的测试数据文件 | 数据文件名 | 数据记录数 | Time参数范围 | | --- | --- | --- | | testing_dc_36_s1.csv | 454 | 1 – 426 | | testing_dc_36_s2.csv | 562 | 2 – 93 | | testing_dc_36_s3.csv | 272 | 2 – 169 | | testing_dc_36_s4.csv | 219 | 2 – 136 | | testing_dc_36_s5.csv | 414 | 4 – 253 | | testing_dc_48_s1.csv | 625 | 4 – 380 | | testing_dc_48_s2.csv | 278 | 27 – 196 | | testing_dc_48_s3.csv | 178 | 12 – 117 | | testing_dc_48_s4.csv | 211 | 5 – 131 | | testing_dc_48_s5.csv | 126 | 4 – 81 | | testing_dc_48_s6.csv | 132 | 2 – 81 | "Init.txt"为配置文件，存储了每次行驶工况前宏簇的初始参数。每个宏簇的初始值设置如下：标识符为"C=0"、车速为"S=00023"、加速度为"A=033"以及方向盘角速度为"SWAV=00359"。