The aim of this project is to develop a robust and accurate model for detecting confidential data left unredacted in medical reports by healthcare professionals. This is crucial for ensuring patient privacy and confidentiality. The project is undertaken by ProvistaAI to address the need for efficient identification of sensitive information within medical documents, including patient names, doctor names, IDs, licenses, clinic names, and location details such as cities, countries, zip codes, and street names.
Can we build models that can predict whether or not data has been anonimized?
Which algorithm - KNN or Random Forests - is more efficient for our purpose of detecting confidential information?
How does the model's performance vary across different confidential data formats, such as patient names, doctor names, patient IDs, and location details?
The dataset comprises human-written radiology reports and synthetic data generated by our team and ProvistaAI to simulate medical reports. The synthetic data mimics real-world medical reports to enhance the model's training process. Key characteristics of the dataset include the variety of confidential data formats present, such as patient and doctor identifiers, clinic names, and location details.
Size of the dataset will be discussed with the TA.
The model class used in this project is supervised learning (text classification). Different algorithms will be used: k-nearest neighbors (KNN) and random forests. These algorithms are chosen for their effectiveness in handling classification tasks and their ability to capture complex patterns in data. This model will be evaluated using metrics such as accuracy, precision, recall, and F1-score.
This simple yet powerful classification algorithm works by finding the k-nearest data points in the training set to the input instance and making predictions based on the labels of those nearest neighbors. It is particularly suitable for this task due to its ability to handle non-linear decision boundaries and adapt to the local structure of the data.
This ensemble learning method constructs a multitude of decision trees during training and outputs the classification of the individual trees. It excels in handling high-dimensional data and mitigating overfitting, making it well-suited for this project's task of detecting confidential information in medical reports.
Pyspark, Python libraries