Project Title: ANALYSING AND IMPROVISING OF MACHINE LEARNING MODELS FOR EARLY DIAGNOSIS OF BRAIN TUMOUR
Mudit Golchha
Dr. Brindha GR, Assistant Professor-II, School of Computing , SASTRA University
Brain tumors, classified broadly as glioma (malignant) and meningeal tumor (non-malignant), present a grave threat by damaging and compressing vital parts of the brain. Traditional diagnostic methods often prove time-consuming, impeding prompt treatment. This project endeavors to expedite tumor detection using machine learning methodologies. Leveraging a dataset featuring metabolite profiles specific to glioma (categorized by grade) and meningeal tumors, ten diverse machine learning models are trained and rigorously tested. These models include Naive Bayes (NB), Generalized Linear Model (GLM), Logistic Regression (LR), Fast Large Margin (FLM), Deep Learning (DL), Decision Tree (DT), Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Machine (SVM), and Evolutionary Heterogeneous Decision Tree (EvoHDTree). Through rigorous evaluation based on performance metrics such as Area Under the Curve (AUC), Accuracy (ACC), and F1-score, these models are compared for their efficacy in discriminating between healthy individuals and those afflicted with carcinoma. Furthermore, the top-performing EvoHDTree model (with an ACC of 0.9195) undergoes enhancement through the novel application of the ADASYN up-sampling algorithm during preprocessing. This augmentation yields the ultimate model capable of distinguishing between malignant and non-malignant tumors with heightened accuracy.
- Developing machine learning methods other than EvoHDTree
- Working on GUI.
- Learnt the impact of upsampled data on various ML models
- Designing GUI.
- Uploading the trained model to GUI
- Overcoming the overfitting of the model.
This project is licensed under the BSD-3 License. See the LICENSE file for details.