(In alphabetical order)
AdultCensusData : Built models using sklearn and analyzed data based on the accuracies of Naive Bayes, Decision Tree, Nearest Neighbour, and Logistic Regression algorithms for better understanding.
Cats & Dogs : Implemented a Convolution Neural Network for classification of images after pre-processing using TensorFlow, which gave a mean test accuracy of 91% across multiple Epochs. Also,
used Transfer learning to train the images on the InceptionV3 model, which gave an average mean test accuracy of 96% for comparison of the results.
DecisionTreenEnsembles : Implementation of the ID3 algorithm and Ensembles(Bagging and Boosting) without libraries using Mutual Information on Monks and
Mushroom dataset along with the implementation of sklearn DecisionTreeClassifier algorithm for result comparison.
ESG Score Generator : Developed a pipeline for computing a statement's correlation with Environment, Social, and Governance using Bert model on embeddings and Faiss search on indexed vactors and further using cosine similarity on ESG Embeddings and Non-ESG Embeddings to calculate the difference to condition if the statement involves ESG factor or not. Used Streamlit to render frontend for input.
Fashion MNIST classification : Implemented DNN model to classify the famous dataset that gave an accuracy of 91.6% on validation.
Human and Horse Image Classification : Trained a CNN model on 500 images of Horses and Humans with a mean accuracy of ~84% on the validation set of around 120 images each.
Next Day Rain Predition : Analyzed the Australian Rain Prediction dataset consisting of 23 features having many missing data and outliers then preprocessed and selected correlated features to train on traditional ML algorithms to evaluate the performance of algorithms by accuracy, scalability, interpretability, robustness to outliers and measure the improvement by introducing Bagging & Boosting.
Object Tracking using Histogram extraction : Used OpenCV to capture the pixel intensity of any object and used the resulting information to track the object.
Picture Stacking for Image Enhancement : Implemented ORB (Oriented and Rotated Brief) and ECC (Enhanced Correlation Coefficient) algorithms on Images
using OpenCV and devised a logical approach to mask dim pixels to further balance quality and noise.
Review Classification using Linear SVM : Used sklearn to extract text features from reviews and trained a model by Pipelining the term frequency-inverse document frequency (tf-idf) Vectorization and training SVM model.
Spam Text Classifier : Basic classification of Email /Texts to Spam or Non-Spam using Pipelining on SVC with accuracy of around 98%.