Enhanced Intracranial Tumor Strain Prediction and Detection Using Transfer and Multilevel Ensemble Learning

Ample amounts of research has been already done when it comes to detection of brain tumors in a computerized way. Considering that the manual approach is highly time consuming and erroneous, digitalizing the whole process can be very helpful and can lead to early detection of brain tumors. And this early detection in turn can minimize the risks since the required treatment can be provided in time. The research previously done is basically focused on machine learning models which are in some cases very weak and can lead to less accurate results. Thus, this approach uses analysis of symptoms well before in advance, transfer learning, ensemble learning, and deep learning methods to tackle the problem. The optimized Densenet model achieved a great accuracy of 93% on training set with 15% loss. It achieved validation accuracy of 90% with loss of 16%. The Resunet model was clubbed with this in the next level of ensemble learning, where it achieved the accuracy of 96% with loss of 10% on training set and 91% accuracy with 14% loss on validation set. The training was done on 60 epochs. Analysis of factors affecting intracranial tumors includes use of Random Forest algorithms that gave 93% accuracy which is higher than other models.