Date of Award


Document Type


Degree Name

Master of Science in Information Systems and Technology


Information and Decision Sciences

First Reader/Committee Chair

Shayo Conrad


This Culminating Experience Project investigated how the densenet-161 model will perform on accident severity prediction compared to proposed methods. The research questions are: (Q1) What is the impact of usage of augmentation techniques on imbalanced datasets? (Q2) How will the hyper parameter tuning affect the model performance? (Q3) How effective is the proposed model compared to existing work? The findings are: Q1. The effectiveness of our model depends on the implementation of augmentation techniques that pay attention to handling imbalanced datasets. Our dataset poses a challenge due to distribution of classes in terms of accident severity. To address this challenge directly we utilize an augmentation process that involves applying transformations to the data. By applying these transformations our aim is to create a training set. This enables our model to grasp and capture the nuances of classes resulting in enhanced prediction accuracy and improved generalization abilities. Q2. Adjusting the settings of algorithms to enhance their performance is an aspect of machine learning known as fine tuning hyperparameters. In one of our experiments, we successfully increased our model's accuracy by 2%, which was quite an improvement. Prior to tweaking the hyperparameters the model only achieved a 90% accuracy rate. This remarkable disparity truly emphasizes the impact that hyperparameter tuning can have on a model's performance. By adjusting these parameters, we were able to unlock the hidden potential of our algorithm and enhance its ability to identify patterns and subtle details within the data. This entire process exemplifies how meticulous fine tuning of hyperparameters can lead to advancements in machine learning outcomes. Q3. The study’s findings show that the current work has achieved an accuracy rate of 88%. However, when we implemented the model, we observed an improvement with accuracy reaching 95%. This increase of 4% is quite notable. Reflects an enhancement in performance. It's clear that the densenet-161 model excels at classifying data, which suggests its effectiveness in applications. This substantial boost in accuracy has ranging implications from improving the reliability of diagnoses to enhancing the precision of image recognition systems. These findings highlight the importance of utilizing models like densenet-161 to achieve levels of accuracy emphasizing their potential for profound advancements in fields reliant on precise data classification and analysis. The conclusions are: Q1. This method can help prevent bias in Favor of the majority class and balance the data. Q2. Hyper parameter tuning helps to improve accuracy. Q3. Densenet-161 model able to achieve a 95% accuracy. Further research topics that our study raises are the prospect of evaluating and training our model with a bigger set of data and fine-tuning other hyperparameters for even greater performance.