A new collaborative learning, called split learning, was recently introduced,\naiming to protect user data privacy without revealing raw input data to a\nserver. It collaboratively runs a deep neural network model where the model is\nsplit into two parts, one for the client and the other for the server.\nTherefore, the server has no direct access to raw data processed at the client.\nUntil now, the split learning is believed to be a promising approach to protect\nthe client's raw data; for example, the client's data was protected in\nhealthcare image applications using 2D convolutional neural network (CNN)\nmodels. However, it is still unclear whether the split learning can be applied\nto other deep learning models, in particular, 1D CNN.\n In this paper, we examine whether split learning can be used to perform\nprivacy-preserving training for 1D CNN models. To answer this, we first design\nand implement an 1D CNN model under split learning and validate its efficacy in\ndetecting heart abnormalities using medical ECG data. We observed that the 1D\nCNN model under split learning can achieve the same accuracy of 98.9\\% like the\noriginal (non-split) model. However, our evaluation demonstrates that split\nlearning may fail to protect the raw data privacy on 1D CNN models. To address\nthe observed privacy leakage in split learning, we adopt two privacy leakage\nmitigation techniques: 1) adding more hidden layers to the client side and 2)\napplying differential privacy. Although those mitigation techniques are helpful\nin reducing privacy leakage, they have a significant impact on model accuracy.\nHence, based on those results, we conclude that split learning alone would not\nbe sufficient to maintain the confidentiality of raw sequential data in 1D CNN\nmodels.\n