Weight prediction for small size object using machine learning and deep learning | |
| Author | Tran Minh Duong |
| Call Number | AIT Project no.PMDS-23-03 |
| Subject(s) | Image reconstruction Deep learning (Machine learning) |
| Note | A project study submitted in partial fulfillment of the requirements for the degree of Professional Master in Data Science and Artificial Intelligence Applications |
| Publisher | Asian Institute of Technology |
| Abstract | In the realm of medical imaging and predictive analytics, this thesis embarks on a comprehensive exploration of five distinct models—VGG, DenseNet, ResNet, Pure CNN, and KNN—aiming to predict small size object mass. The primary objective is to evaluate their performance and discern their individual strengths and weaknesses. This research venture spans diverse convolutional neural network architectures, traditional machine learning, and a hybrid approach. The VGG model demonstrates robust predictive capabilities, showcasing rapid convergence and achieving a low Mean Absolute Error (MAE) of 12.84. In contrast, DenseNet's slower convergence and higher MAE of 27.94 suggest potential avenues for improvement, prompting consideration for architectural adjustments or extended training. ResNet, while displaying diminishing training loss, faces challenges with validation loss escalation, resulting in a higher MAE of 138.26. The simplicity of the Pure CNN model surprises with competitive performance, reaching a final MAE of 14.69. KNN, a non-neural network approach, provides quick and commendable results with an MAE of 14.21. This thesis delves into the convergence patterns, losses, and evaluation metrics of each model, offering a detailed examination of their performance. Comparisons highlight the predictive prowess of VGG, the nuances of DenseNet, challenges faced by ResNet, efficiency of Pure CNN, and the commendable speed of KNN. Unexpected findings add depth to the analysis, such as ResNet's struggle with validation loss and the surprising efficiency of Pure CNN. Looking forward, the exploration of these models brings forth key insights and potential directions for future research. The nuanced understanding gained from this study provides a foundation for optimizing model selection based on specific task requirements, refining architectures, and exploring advancements in predictive modeling for mammographic breast density. |
| Year | 2023 |
| Type | Project |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Professional Master in Data Science and Artificial Intelligence Applications (PMDS) |
| Chairperson(s) | Chaklam Silpasuwanchai |
| Examination Committee(s) | Vatcharaporn Esichaikul;Chantri Polprasert |
| Degree | Professional Master in Data Science and Artificial Intelligence Applications - Asian Institute of Technology, 2023 |