Crash Detection Using an IOT Based Sensor and Health Related Features

DOI : 10.17577/ICCIDT2K23-204

Download Full-Text PDF Cite this Publication

Text Only Version

Crash Detection Using an IOT Based Sensor and Health Related Features

ICCIDT – 2023 ConfereInScSeNP: r2o2c7e8e-d0i1n8g1s

ICCIDT – 2023 Conference Proceedings

Abin Moses Andrews1

1Student, Dept. Of Computer Science & Engineering, Mangalam College of Engineering, India,

G Harikrishna3

2 Student, Dept. Of Computer Science & Engineering, Mangalam College of Engineering, India,

Asin Tomy2

3Student, Dept. Of Computer Science & Engineering, Mangalam College of Engineering, India,

Gayathri R Krishna 5

5Assistant Professor, Dept. Of Computer Science & Engineering, Mangalam College of Engineering, India,

Abstract There are a lot of people facing a problem of unexpected death or health damage due to the lack of medical care at the right time, especially elderly people, patient with disabilities and people that are living alone who are required to be continuously under surveillance for the purpose of safety and emergency response. Most of the works done in this field imposed the restriction of fixing the smartphone in certain position on human body to easy infer the emergence case from the data of the smartphone sensors. To overcome this restriction, the proposed system incorporated a smartwatch, together with smartphone freely carried by the user. for better performance results. The use of smartwatch assisted in providing distinct separable signal variation from the smartwatch accelerometer and gyroscope sensors to recognize emergency case such as falling, car accident and heart rate failure. Immediately after cases that are mentioned previously the proposed systems ends details information such as videos, location, heart rate etc. to the emergency centre and emergency contact to provide help at the right time. The system was practically tested in real simulated environment and achieved quite very good performance results. The chatbot will provide quick answers to FAQs by setting up rule-based keyword chatbots with ¨if/then¨ logic. This chatbot will use a series of well-defined rules to guide customers through a series of menu options that can help answer their questions. It will be there for customers 24/7 on their preferred channels, and simultaneously handle more queries at once.

KeywordsInternet of Things,Arduino.



    on the Internet, especially on social media, it is becoming increasingly important to identify and potentially prevent the transmission of hate speech, i.e. fight against racism and sexism. With the vast amount of user-generated information on the Internet, especially on social media, it is becoming increasingly important to identify and potentially limit the spread of hate speech. Hate speech and defamatory comments against another person's religion, ethnic origin, or sexual orientation are prohibited by law. In many countries, anyone who incites violence or genocide is considered a criminal. In addition, many governments prohibit the use of symbols of totalitarianism and restrict freedom of assembly in the case of fascism or communism. However, not everyone has equal access to this public space and not everyone has the right to express themselves without fear. Hostile and disrespectful communication on the Internet drowns out the voices of marginalized and underrepresented groups in the public conversation. This helps us to understand this and mitigate .

    Hate speech on the Internet and social media not only causes friction between groups of people, but it can also cause harm businesses and cause really important problems. For these reasons, websites such as Facebook, YouTube, and Twitter are limited hate speech. However, tracking and filtering all content always problems. For this reason, many tests have been conducted to learn how to automatically detect hate speech. Most of this hate speech detection work attempts to create dictionaries of hate phrases and expressions or categorize hate speech into two categories: "hate" and "don't hate". However, assessing whether a sentence contains hateful content is always difficult, especially when hate speech is masked under sarcasm or when hate is not clearly expressed. race or prejudice. The goal of this study was to extract hate

    speech from social media content in an online forum. We have

    Volume 11, Issue 01

    Published by, wpwrwop.iojesretd.oraghate speech visualization and recognition system based on the deep attention technique. In a study of online

    trends, users

    and videos. Given the large amount of user-generated content communicated hate speech in response to difficuISltSieNs: i2n2t7h8e-0181

    ICCIDT – 2023 Conference Proceedings

    Communication or an online system (especially during a pandemic).


    Deep learning techniques have proven to be very effective in classifying hate speech. The performance of deep learning- based approaches has outperformed classical machine learning techniques such as support vector machines (SVMs), gradient- enhanced decision trees (GBDTs), and logistic regression. Among the deep learning-based classifiers, convolutional neural networks (CNNs) record local patterns in the text. A long-term memory model (LSTM) based on a Deep Learning Model or Gated Recurrent Unit (GRU) captures long-range


    [1] Deep learning to detect hate speech in tweets: Detecting hate speech on Twitter is essential for applications such as extracting controversial events, creating AI chatbots, content recommendations, and sentiment analysis. We define this task as being able to classify a tweet as racist, sexist or not. The complexity of natural language constructs makes this task very difficult. We perform extensive tests with several deep learning architectures to learn how to embed semantic words to address this complexity. Our tests on a benchmark dataset of 16,000 annotated tweets show that such deep learning methods outperform modern character/word n-gram methods by about 18 F1. With the dramatic increase in social interactions on online social networks, there has also been an increase in hate activity aimed at exploiting these infrastructures. On Twitter, hate tweets are tweets containing abusive language aimed at individuals (online followers, politicians, celebrities, products) or specific groups (a country, LGBT, etc.) religion, gender, organization, etc.). Such hate speech detection is important to analyze the overall sentiment of one group of users towards another and to prevent related illegal activities.

    [2] CNN on hate speech and identifying offensive content in Hindi: describes the best group solution for task 1 for Hindi in HASOC competition organized by FIRE 2019. The mission is to identify hate speech and offensive language in Hindi. Specifically, it's a binary classification problem where a system has to classify tweets into two classes:

    (a) hateful and offensive (HOF) and (b) not hateful or offensive (NOT). Contrary to the popular idea of pre-training word vectors (aka word embedding) with a large corpus of corpus from a common domain such as Wikipedia, here we have used a relative collection small relevant tweets (i.e. random and sarcastic tweets in Hindi and Hinglish) to practice in advance. Here, they trained a convolutional neural network (CNN) on pre-trained word vectors. This approach allows us to be ranked first for this mission out of all the teams. otherwise it is labeled as NO. There has been significant research on hate speech and the identification of offensive content in several languages, particularly in English [3,2,6,25,24]. However, there is alack of work in most other languages. The proposed method is based on very little preprocessing and feature engineering compared to many existing methods. [3] Development of an online hate classifier for multiple social media platforms: The growth of social media allows people to express their emotions and also feelings

    At the same time, however, it leads to the emergence of conflict and hatred, making the online environment unattractive for users. Although researchers have found that hate is a cross- platform problem, there is a lack of online hate detection models that use cross-platform data. To fill this research gap, we are collecting a total of 197,566 reviews from four platforms: YouTube, Reddit, Wikipedia and Twitter, with 80% of comments labeled as non-hate and the remaining 20% labeled as hateful. Then we test some classification algorithms (Logistic Regression, Naïve Bayes, Support Vector Machine, XGBoost and Neural Network) and feature representation (Bag-of-Words, TF-IDF, Word2Vec, BERT and their combinations). Although all models significantly outperformed the benchmark keyword-based classifier, XGBoost using all the features would perform best (F1 = 0.92). Feature importance analysis indicated that BERT features had the most impact on predictions.

    [4] Hate me don't hate me: Detect hate speech on Facebook: While promoting communication and facilitating information sharing, social networking sites are also used to launch harmful campaigns against specific groups and individuals. Cyberbullying, inciting self-harm, sexual assault are just some of the serious effects of large-scale online attacks. In addition, attacks can be made against groups of victims and can escalate into physical violence. In this work, we aim to prevent and stop the alarming spread of such hate campaigns. Using Facebook as a benchmark, we looked at the text content of comments that appeared on a set of Italian public pages. First, we introduce multiple hate categories to distinguish the type of hate. Discovered comments are then annotated by up to five separate annotators, depending on the identified taxonomy. Leveraging the features of syntactic morphology, affective polarization, and word-integrated lexicon, we design and implement two classifiers for the Italian language, based on different learning algorithms: the first is based on support vector machines (SVM) and the second is based on a specific recurrent neural network called Long Short Term Memory (LSTM).

    [5] Hate Speech Detection: A problem solved? Long Tail's hard case on Twitter: This work makes several contributions to the state of the art in this field of study. First, in-depth data analysis to understand the extremely imbalanced nature and lack of discriminatory characteristics of hate speech in the typical datasets one faces in tasks. such service. Second, new DNN-based methods are proposed for such tasks, specifically designed to capture latent features that are potentially useful for classification. Finally, the methods have been carefully evaluated on Twitter's largest data collection of hate speech, to show that they can be particularly effective at detecting and classifying content. hate speech (as opposed to non-hateful content) that we have shown is more effective than hard and arguably more important in practice. The end results set a new standard in this field of research. With the growing popularity of deep learning-based NLP models, interpretable systems are needed. But what is interpretability and what constitutes interpretability? In this opinion piece, we reflect on the current state of research on interpreting assessment.


    1. Proposed System III

      To build a hate speech detection model using LSTM, we can follow these general steps:

      1. Data Collection: Collect a large dataset of labeled data that contains examples of hate speech and non-hate speech. There are several publicly available datasets for hate speech detection that can be used.

      2. Data Preprocessing: Preprocess the data by removing irrelevant information such as stopwords, punctuation, and special characters. Then tokenize the text and convert it to a sequence of integers to feed into the LSTM model.

      3. Word Embeddings: Use pre-trained word embeddings such as GloVe, FastText or Word2Vec to represent each token in the text with a dense vector. This will help the LSTM model to learn better semantic relationships between words.

      4. LSTM Model Architecture: Define the LSTM model architecture with an embedding layer followed by one or more LSTM layers. The output of the LSTM layers will be fed into a fully connected layer with a sigmoid activation function to produce a binary classification output.

      5. Training: Train the LSTM model on the preprocessed and embedded data using the back propagation algorithm with cross-entropy loss. Adjust hyper parameters such as learning rate, batch size, and number of epochs to optimize the model's performance.

      6. Evaluation: Evaluate the performance of the LSTM model on a separate test set using metrics such as accuracy, precision, recall, and F1 score. Tweak the model parameters and architecture as needed to improve its performance.

      7. Deployment: Deploy the trained LSTM model as a service or integrate it into an application for real-time hate speech detection.

        It is also important to note that hate speech detection is a complex and nuanced problem, and it may be necessary to incorporate additional techniques such as topic modeling, sentiment analysis, and user profiling to improve the accuracy of the model.

    2. Algorithm

      1. Data preparation: Collect and prepare a dataset of text labeled as either hateful or non-hateful. This dataset should be large enough to train the algorithm.

      2. Tokenization: Convert the text into a numerical format that can be processed by the algorithm. Tokenization involves breaking up the text into words or subwords, and assigning each token a unique numerical value.

      3. Embedding: Transform the tokens into a dense numerical representation using a word embedding technique. This step captures the semantic relationships between words and their contexts.

      4. Bidirectional LSTM: Train a bidirectional LSTM neural network to classify the text as hateful or non-hateful.

      1. Bidirectional LSTMs process the text in both forward and backward directions, allowing them to capture long-term dependencies in the text.

      2. Output layer: The output layer of the LSTM network is a binary classifier that outputs a probability of the text being hateful or not.

      3. Evaluation: Evaluate the performance of the algorithm on a test dataset. Common evaluation metrics include accuracy, precision, recall, and F1 score.

      4. Tuning: Fine-tune the hyperparameters of the algorithm to optimize performance. This can include adjusting the number of LSTM layers, the number of neurons in each layer, and the learning rate.

      5. Deployment: Deploy the trained model to classify new text as hateful or not.

      Overall, using a bidirectional LSTM algorithm to detect hate speech involves a combination of natural language processing techniques and machine learning algorithms. It is important to use a large and diverse dataset to train the algorithm and to carefully evaluate its performance to ensure its effectiveness.

    3. System Architecture

      Fig.1.system architechture

    4. DFD LEVEL 0




    have important implications for content monitoring and moderation.

    online. Con

    speech detect contextualizati understand th of language techniques ca subtle nuanc conventional algorithms mi contextual mo distinguish sarcasm, sarca

    textualization: Hate ion can benefit from on to better e underlying meaning

    . Contextualization n be used to identify es in language that hate speech detection ght miss. For example, dels can be trained to hate speech from sm, or humour. AI can

    explain: Explainable AI can help improve the transparency and accountability of hate speech detection models. It can be used to identify features and patterns in data that contribute to hate speech detection. This will allow researchers and moderators to better understand the model's decision-making process and make improvements accordingly. Opponent attacks: Adversarial attacks can be used to manipulate hate speech detection models by intentionally introducing errors into the data. Therefore, future research in this area can focus on developing robust

    The proposed model aims to reduce the number of data

    annotation operations. Therefore, this technique contributes generalization of the apprenticeship system. Word-classified semantic vectors combine word information the context in which they occur. From the combined result uses semantic information to help select a subset of unlabeled text. This approach identifies unlabeled text-based cases of active learning. Method of integrating new learning points into model training. Hate speech detection using a BiLSTM (Bidirectional Long Short-Term Memory) model will typically be a binary classifier of whether a given text should be considered hate speech. The output is usually a probability score between 0 and 1, where 0 indicates that the text is not hate speech and 1 indicates that the text is hate speech. The BiLSTM model will be trained on a dataset of labeled examples of hate speech and non-hate text, using techniques such as word embedding, recurrent neural networks, and annotation mechanisms. idea. The model will then be used to predict whether new unseen texts are hate speech. It should be noted that the accuracy of hate speech detection using the BiLSTM model (or any other machine learning model) can vary depending on the quality and variety of the training data. , as well as the complexity and efficiency of the model 's architecture and parameters . In addition, determining what constitutes hate speech can be subjective and context dependent, so there may be some degree of ambiguity or disagreement in the labeling of some texts.


    As natural language processing (NLP) and machine learning continue to evolve. BiLSTM (Bidirectional Long Short-Term Memory) is a deep learning algorithm that has been proven effective in text classification tasks such as sentiment analysis and hate speech detection. Some potential future directions for research and development in this area include: Multilingual hate speech detection: BiLSTM can be trained to detect the hate speech in multiple languages, which is especially important due to the global nature of social media and the internet. As a result, researchers can develop models capable of detecting hate speech in a variety of languages, which will

    detection using BiLSTM is very bright and there are many opportunities for research and development in this area. As natural language processing and machine learning continue to evolve, so will these models' ability to identify and moderate online hate speech.


    In daily life, as the use of social media increases, people seem to think that they can say or write whatever they want. Due to this reflection, hate speech has increased, so there is a need to automate the process of classifying hate speech data. Interpretable natural language processing and deep learning have been adopted in recent years. Existing models are based on static data. Therefore, most traditional algorithms cannot account for significant changes. The proposed supervised learning method first labels the text and then trains the model. Two-way LSTM can achieve excellent accuracy when combined with active learning and attention networks.


The authors would like to thank Principal Vinodh P Vijayan, Neethu Mariya John, H.O.D, Faculty of Computer Science, for their appropriate guidance, valuable assistance and helpful comments during the proofreading process.


[1] Deep Explainable Hate Speech Active Learning on Social-Media Data,Usman Ahmed and Jerry Chun-Wei Lin Senior Member, IEEE

[2] A. Rajkomar et al., Scalable and accurate deep learning with electronic health records, NPJ Digit. Med., vol. 1, no. 1, pp. 110, 2018.

[3] K. W. Johnson et al., Artificial intelligence in cardiology, J. Amer. College Cardiol., vol. 71, pp. 2668 2679, Jun. 2018.

IICCCCIIDDTT–22002232CCoonnffeerreennccee PPrroocceeeeddiinnggss

[4] C. Krittanawong, H. Zhang, Z. Wang, M. Aydar, and T. Kitai, Artificial intelligence in precision cardiovascular medicine, J. Amer. College Cardiol., vol. 69, no. 21, pp. 26572664, 2017.

[5] E. Choi, M. T. Bahadori, A. Schuetz, W. F. Stewart, and

J. Sun, Doctor AI: Predicting clinical events via recurrent neural networks, in Proc. 1st Mach. Learn. Healthcare Conf., vol. 56, Aug. 2016, pp. 301318.

[6] R. C. Feldman, E. Aldana, and K. Stein, Artificial intelligence in the health care space: How we can trust what we cannot know, Stan. L. Poly Rev., vol. 30, p. 399, Jul. 2019.

[7] D. Gunning, Explainable artificial intelligence,

Defense Adv. Res. Projects Agency (DARPA), p. 2, 2017. [8] A. Holzinger, G. Langs, H. Denk, K. Zatloukal, and H.

Müller, Caus- ability and explainability of artificial intelligence in medicine, WIREs, Data Mining Knowl. Discovery, vol. 9, no. 4, p. e1312, Jul. 2019.

[9] A. Vellido, The importance of interpretability and visualization in machine learning for applications in medicine and health care, Neural Comput. Appl., vol. 32, pp. 1806918083, Feb. 2019.

[10] S. L. James et al., Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990 2017: A systematic analysis for the global burden of disease study 2017, Lancet, vol. 392, no. 10159, pp. 17891858,

Nov. 2018, doi: 10.1016/S0140- 6736(18)32279-7.

IISSSSNN:: 22227788–00118811

IICCCCIIDDTT–22002232CCoonnffeerreennccee PPrroocceeeeddiinnggss