In this thesis, three generative models are compared for the task of weather based music generation; Conditional Variational Autoencoder (cVAE), a Conditional Generative Adversarial Network (cGAN) and Long Short-Term Memory (LSTM) network. Mentioned models are implemented and trained on combination of large MIDI corpus with historical weather information.
Performance of models in this paper are evaluated using metrics that capture musical diversity and quality, such as pitch range, unique pitches and pitch variance, and for fidelity to real data is measured by mean squared error and KL divergence. Results of the paper showed that the cVAE produced the most diverse music for music that is context sensitive. Building upon these findings, the thesis presents the architecture and functionality of a real time full stack application. This system acquires weather data, processes it through the cVAE and generates musical compositions. These compositions are then delivered and rendered through a web interface. This research shows the potential of combining environmental data with AI generated music and gives a framework for applications such as adaptive game soundtracks, mood-based music therapy, or dynamic background music systems that respond to the user's environment. Thesis also points out limitations of the models and gives future research directions in terms of hybrid architectures, richer environmental data representations, and user perception studies.]]> Enterprise Resource Planning (ERP) systems are of immense importance in simplifying business operations. However, most ERP projects fail owing to the complexity and scope of the projects. The present research attempted to determine the outcomes of ERP projects by employing machine learning methods and addressing factors which determine whether projects fail or succeed. This dissertation obtained data from different aspects of the projects that included successful and unsuccessful ERP deployments in terms of within which industry, project magnitude, the level of budget and time exceeding, background of team experience as well as technical challenges faced amongst others.
The research includes machine learning methods such as logistic regression, decision trees, and random forests in order to assess the importance of the relevant predictors of any project. By training and testing these applications on a sample composed of both successful and non-successful ERP projects, the goal of the model is to seek for factors and patterns which could help in forecasting troubling tendencies. This research is aimed at devising a functional framework that can be used by project managers, enabling them to take action before issuing their project plans for ERP systems. Such a predictive model could significantly help in decreasing the rates of ERP failures and hence assist businesses in carrying out successful implementations and enhancing their returns on technology investment.]]> Heart strokes remain one of the leading health risks in the world today. Timely prediction can significantly improve patient outcomes and healthcare resource allocation. This study aims to harness machine learning techniques to develop efficient predictive models for the early detection of heart strokes.
Research is based on a dataset created by combining different (five) datasets. The dataset encompasses patient demographics, clinical measurements, and historical medical records. The analysis focused on five machine learning models: Logistic Regression, Decision Tree, K-Nearest Neighbors, Random Forest, and Support Vector Machine.
The goal was not only to test different algorithms, but also to understand how data preparation, feature selection, and model choice impact the final results. The models were trained and tested on both the original dataset and an extended version, where new features were added by combining existing ones.
The results showed that models such as Logistic Regression, Decision Tree, and KNN performed better when applied to the original data. The Decision Tree model achieved an accuracy of 87.8% and an F1 score of 0.881, while Logistic Regression and K-Nearest Neighbors each attained F1 scores of 0.850 and 0.849, respectively. On the other hand, Random Forest and SVM showed significant improvements with the extended dataset. Random Forest performed the best overall, with an F1 score of 0.920 and an accuracy of 91.6% with enhanced results.
SVM also benefited from enhanced results, improving its F1 score from 0.892 to 0.879, which highlights how specific models can leverage additional features for improved generalization.
This tool could help detect risks earlier, allowing for timely interventions and prevention, thereby reducing the burden of strokes on healthcare systems and improving patient care. Limitations include data quality and availability, as well as potential bias in healthcare records.]]> Diabetes is a growing global health issue, and early prediction is key to preventing its effects. This thesis develops predictive models for diabetes using various machine learning methods, including Logistic Regression, Decision Trees, K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and XGBoost, using the Diabetes Health Indicators dataset, which covers clinical, lifestyle, and demographic factors. Feature selection identifies the most important diabetic predictors, and model performance is evaluated using macro average and weighted average metrics, accuracy, precision, recall, F1-score, and error metrics (MSE and RMSE) to provide a thorough evaluation of model performance across the classes. Both SVM and Random Forest performed best overall, with an accuracy of 0.86. They also performed exceptionally well in weighted average and macro average measures, with overall recall and F1-scores of 0.86. SVM has the highest precision performance at 0.88, with Random Forest achieving the next best score of 0.87. These models are very dependable for diabetes prediction tasks because of their remarkable balance while handling both classes. SVM and Random Forest offer more dependable performance on a range of metrics, as evidenced by the weaker outcomes of Decision Tree, KNN, XGBoost, and Logistic Regression.

]]>