Volume 12 | Issue 12 |

 Abstract

Driven by exhaustion is one of the leading causes of road accidents across the world which result in many reported deaths each year. This project attempts to mitigate the challenge using machine learning and deep learning techniques for easier and precise detection of fatigue. EEG signals and video images from the DROZY dataset are utilized to train ML and DL models, respectively. SVM, Random Forest, and KNN are some of the machine learning algorithms utilized for the analysis of EEG signals while the images are analyzed through CNN, ConvLSTM, and also a multi-model approach CNN+ConvLSTM. Further accuracy is enhanced by employing ensemble techniques such as Bagging Classifier. PV and PCA methods were applied to obtain superior model performance resulting in all 100% accuracy for all CNN-based approaches. Thus, this system provides strong approaches toward supporting real time driver fatigue assessment.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

ConvLSTM, EEG, KNN

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The emergence of exotic diseases points out a pressing void for swift medical aid. In this project, the authors focus on the recommendation of drugs utilizing a combination of patient reviews, disease description and machine learning aided sentiment analysis. The methodology incorporates TF-IDF, Bag of Words, and Word2Vec for feature extraction along with Logistic Regression and Multilayer Perceptron (MLP) as algorithms. For training and testing of the models the authors utilized the DRUGREVIEW dataset publicly available in UCI, reviews and drugs with ratings are used to anticipate drug outcomes. It was determined that for predicting drug outcomes MLP shows greater efficiency than the rest, hence it was chosen as the core algorithm in the developed system. Additionally doctors' prescribed medications are supplemented with suggested sentiments so that the chances of self medicine withdrawal are decreased and patients make better choices with regard to the medications.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

MLP, TF-IDF, Logistic Regression

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Diagnosing heart disease is a medical problem that is both critical and has a time factor which has to be taken into consideration in order to be treated. This project looks into the performance analysis of four machine learning models which are SVM, KNN, Logistic Regression and XGBoost, with and without parameter tuning via GridSearchCV. The analysis set up is based on Hungarians Cleveland data set which has attributes for predicting whether the patient is likely to have heart problems. XGBoost computes higher accuracy among the algorithms but had prolonged computation time. The study therefore extends with Random Forest in that regard which parallels the accuracy of XGboost but decreased computation time. This project underlines the importance of parameter tuning in the improvement of model performance and identifies Random Forest as a low cost method which will result in faster and more accurate predictions for heart diseases.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

XGBoost, KNN, SVM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper describes a professional system designed to detect cases of ransomware assaults based on data related to processor usage and disk usage. System processes and the activities dealing with files are some of the classical methods used, but at times power effectiveness is compromised and the methods are not very reliable. In order to compensate for these, VMware environment is used in this work in order to obtain HPC as well as I/O events without degrading performance. The machine learning algorithms that were employed in the evaluation of the model included SVM, Random Forest, and XGBoost, where Random Forest and XGBoost achieved 98% accuracy. In addition, other DNN and LSTM deep learning models were applied, and an extension with CNN2D was reported with most accuracy of 98.83%. This self-learning system of detection has changed the way ransomware detection is done without compromising the performance of the system in a big way and is an effective means of dealing with cyber threats.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

CNN2D, DNN, LSTM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

CNN, ANN, SVM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Today almost every person uses a credit card, but fraudulent activities are a great concern to both the customers and the financial institutions. This project compares the performance of several machine learning algorithms such as Decision Tree, KNN, Logistic Regression, SVM, Random Forest, and also XGBOOST, as well as a combined approach, which introduces the use of deep learning CNN. The main issue tackled is the problem on dataset which is fairly skewed, and hence, the normal transactions are a majority, while the fraudulent transactions are few. PCA for feature selection and SMOTE for data balancing techniques are applied for this purpose. A combination of the two, wherein CNN is combined with Decision Tree, increases all detection accuracy to 100%. This project offers valuable contributions as it highlights sample solutions to the problem of credit card fraud by using the Canadian Credit Card Dataset in a fast and accurate way.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

CNN, XGBOOST, SVM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The accurate prediction of crop is crucial for effective agricultural planning and resource mobilization. This project constructs and enhances the system of predicting crop yield harnessing the features of the agriculture environment. The important environmental features to be focused on includes among other soil type, rainfall and temperature by implementing various feature selection methods including BORUTA and Recursive Feature Elimination (RFE). These features are fed to the ensemble of machine learning algorithms such as Random Forest, SVM and KNN to improve the prediction accuracy. The system also provides fertilizer application and yield prediction. Testing in databases yields promising results in the improvement of precision and decision making. The proposed model proves to be reliable and cost efficient and assists farmers with actionable information to enhance crop productivity and sustainable farming techniques.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

BORUTA, SVM, KNN

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

In this work, we propose an ANN-based AI cyber security framework for detecting cyber attacks. By converting security incidents into single entities, the architecture takes advantage of advanced learning models including CNN and LSTM to improve the detection rate of the system. The AI-SIEM system decreases the occurrence of false alarms, thus enabling more efficient and faster response to changing cyber attacks. Experiments carried out on benchmark datasets, such as NSLKDD, CISIDS2017, results high values when compared to other machine learning techniques like SVM, k-NN and Decision Trees. The technique in question has been designed with an emphasis on practical scenarios, where issues such as data annotation and overfitting are expected. The results confirm that the elaborated framework is competent enough for intrusion detection, working with large volumes of information and responding to a changing threat landscape, thus providing strong defense in various spheres of cybersecurity.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

LSTM, k-NN, SVM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This project applies deep transfer learning to understand facial features and ascertain certain diseases including beta-thalassemia, hyperthyroidism, Down syndrome and leprosy. System incorporates VGG16, ALEXNET and Kernel SVM as advanced models and the Disease Specific Face dataset for training and testing performance evaluations. DLIB along with GABOR is used for the preprocessing to extract features and for facial alignment respectively. Of all the models, VGG16 and LSTM had performed best with accuracy of about 99%. The system also uses other performance metrics including accuracy, precision, recall and F-score to confirm the importance of working systems within the provided scope. It can be concluded based on the results of the project that the combination of transfer learning and feature extraction improves the medical diagnosis process and provides an effective overall scheme of facial recognition-based disease detection system.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

DLIB, GABOR, VGG16

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

NLP, SVM, SQP

  License

Creative Commons Attribution 4.0 and The Open Definition