SAGNIK GHOSAL

• Fraudulent Medical Claims Detection                                                   [Github Link]

  • ​Developed an infrastructure to detect false claims using data visualization and modeling. Used PyOD library to implement supervised and unsupervised models and Flask to develop the web application.                                        

​

• Anomaly Detection in Streaming Environment                                            [Github Link]

  • ​Developed a dynamic tree clustering-based anomaly detection model for streaming environment. The proposed model is fast, memory efficient, highly accurate, and requires minimal user input. Tested on 31 real-life and 22 synthetic datasets, it surpasses the performance of existing state-of-the-art streaming anomaly detection frameworks.   

​

• Human Activity Recognition                                                            [Github Link]

  • Developed a fuzzy ensemble-based approach using deep neural networks for Human Activity Recognition using sensor multimodal data. Proposed a novel rewarding and penalization logic that improved the existing state-of-the-art accuracy by 5% while reducing the computation time by 10%, tested on 4 publicly available real-life HAR datasets.        

​

• sHEMO: Smartphone Spectroscopy for Blood Hemoglobin Level Monitoring                 [Github Link]

  • ​​Developed an algorithm in Python that predicts blood hemoglobin with an accuracy of 97% and anemia screening with an accuracy of 89%, providing an economical option for mass anemia screening. Published the findings in IEEE Sensors journal.                                        

​

• gluCam: Smartphone Based Blood Glucose Monitoring and Diabetic Sensing               [Github Link]

  • ​Led a 4 member team to develop a smartphone-based blood glucose measurement and diabetic sensing model that reported an accuracy of 91.89% and sensitivity of 94% in diabetic sensing tested on 81 patients.                                                                                          

• Chatbot                                                                                [Github Link]

  • Developed a chat application using Python and Google Cloud Console by performing data pre-processing, vectorizations, and implementing an encoder-decoder-based LSTM neural network using PyTorch. The model reported an overall accuracy of 93.5% on the test dataset. It reduced customer waiting time and facilitated easier ordering, increasing sales in the restaurant where it was deployed.                                                                                           

• Non-Invasive Anemia-Polycythemia Detection in the IoMT                               [Github Link]​

  • ​Developed an algorithm in Python that estimates blood hemoglobin using both conjunctiva and fingernails as ROI by leveraging the feature of the IoMT. The model reported an accuracy of 98.4% in hemoglobin prediction, tested over 100 patients.

​ 

• Object Detection and Distance Estimation                                              [Github Link]

  • ​Developed an algorithm in MATLAB that estimates the distance of the object(s), from a two-camera setup using the triangularization technique. The model reported an average accuracy of 95% over 50 images, with the highest being 99.8%.   

​

• Depth Map From Stereoscopic Images                                                   [Github Link]

  • ​Successfully implemented state-of-the-art depth estimation technique using MATLAB programming language. The model estimates disparity and re-constructs the point cloud using stereo images.

​​

• Comparison of Filtering Techniques                                                   [Github Link]

  • ​Developed a comparison algorithm in Python that demonstrated the efficacy of a Gaussian filter in the case of Gaussian noise and a Median filter for salt and pepper noise.