First Year Project
Skills Gained
Predicting a skin lesion’s state
Data Preprocessing & Image Segmentation
Machine Learning Modeling (KNN, Logistic Regression, Nearest Centroid, Random Forest)
Feature Extraction (Asymmetry & Color)
Model Evaluation & Hyperparameter Tuning (Cross-validation, Accuracy, F1-score, and AUC-ROC)
Critical Analysis & Research Writing
Machine Learning Modeling (KNN, Logistic Regression, Nearest Centroid, Random Forest)
Feature Extraction (Asymmetry & Color)
Model Evaluation & Hyperparameter Tuning (Cross-validation, Accuracy, F1-score, and AUC-ROC)
Critical Analysis & Research Writing
2023
In this project, our team developed a machine learning models to classify skin lesions, focusing on identifying types linked to skin cancer. Using the PAD-UFES-20 dataset, which includes clinical images and metadata of various skin lesions, we preprocessed images with Multi-Otsu thresholding and Snake Contour segmentation to isolate key features like color, asymmetry, and texture. We tested models including K-Nearest Neighbors, Nearest Centroid, Logistic Regression, and Random Forest. Through grid and random search, we refined model performance, using metrics like accuracy, F1-score, and AUC-ROC to evaluate robustness and reliability. The final model shows potential as a clinical diagnostic tool, achieving optimal accuracy with Random Forest after tuning.
In this project, our team developed a machine learning models to classify skin lesions, focusing on identifying types linked to skin cancer. Using the PAD-UFES-20 dataset, which includes clinical images and metadata of various skin lesions, we preprocessed images with Multi-Otsu thresholding and Snake Contour segmentation to isolate key features like color, asymmetry, and texture. We tested models including K-Nearest Neighbors, Nearest Centroid, Logistic Regression, and Random Forest. Through grid and random search, we refined model performance, using metrics like accuracy, F1-score, and AUC-ROC to evaluate robustness and reliability. The final model shows potential as a clinical diagnostic tool, achieving optimal accuracy with Random Forest after tuning.