As part of a research project, I studied the impact of different image filters on face detection accuracy using various machine learning algorithms. We created custom datasets with both unfiltered and filtered images (using blur, noise, mosaic, cut-out, and shear). Six algorithms including CNN, KNN, SVM, and Random Forest were tested. The results showed that most filters had little to no effect on accuracy, except for the blur filter, which significantly impacted detection performance. Statistical analysis using paired t-tests supported these findings.
Paper Link →In the present paper, a comparative study for different feature selection methods like correlation, mutual information, Fisher's test etc. has been made. Six datasets have been considered and the three feature selection techniques have been applied to these datasets. Selected features from different datasets using three different feature selection methods have been analyzed and further performance of five different classification algorithms have been examined for these features. Kurskal Wallis test is applied to check whether the performance of these classification algorithms for the selected features using three different feature selection methods for different datasets is the same or not.
Paper Link →In this paper, I have worked on to improve the overall accuracy and the individual accuracies of the given gestures before developing the sign language gestures prediction system. I took a dataset of pre-processed images from Kaggle and further applied CNN model with four layers of relu activation function, three layers of pooling and a layer of softmax activation and after execution I got an overall accuracy of 99.95%.
Paper Link →Click below to connect and talk to the agent.