- Creation of Complex Percussion Patterns in Indian Classical Music using Lasso Regression Learner. (In progress)
2. Determination of Age of an Abalone. (completed)
The age of an abalone is to be determined based on few measurements made on the abalone. The Classication framework here was a pipeline of PCA, Kmeans, Sparse Representation Learner, and SVM. An accuracy of 33% was achieved, to my knowledge this is the highest accuracy achieved on this 29 class dataset.
3. Offline Tamil writer verification via Bagged Classification trees. (completed).
. The algorithm is trained with handwritings of different subjects. Here given a handwriting sample the algorithm finds out which handwriting the sample belongs to. Classification trees are used and are ensembled via bagging. The effect of number of trees on the prediction accuracy was studied.
4. Protein Localization Determination for Ecoli Bacteria
The objective of the project was to determine sites of localization of proteins, here it was framed as a 8 class classification problem. An accuracy of 91% was achieved using UCI college repository, the accuracy achieved is one of the highest reported accuracies.
5. Tamil offline handwritten OCR via Sparse Representation learners ensembled by SVM. (completed)
In this project different Lasso regression models are ensembled using SVM. This ensembled unit is the classifier used in the project. After experimentation and statistical analysis the apt feature extraction methods that described the Tamil characters were chosen. The accuracy of the algorithm compared favorably with the state of the art.
6. Performance comparison of CART, naive bayes, SVM on the TAG-ME IISc dataset. (completed)
The objective was study the different performances of CART, naive bayes, and SVM. It was observed that CART which was an emsembled learning model, performed the best. It achieved 81% on the tag-me dataset.
7. CLOS network simulator. (completed)
A scalable program was written to simulate the CLOS network. This was used to investigate the effectiveness of the CLOS network in a multi-processor environment.