Greater Munich Metropolitan Area
Driving digital transformation in the Healthcare and Life Sciences sector to improve patient outcomes. Soundarya is a Sr Industry Advisor at SAS, where she supports the Global Health and Life Sciences Practice with her expertise in machine learning, text mining, and predictive modeling. She has over a decade of experience in the field of computer science, with a focus on regulatory data science. In her current role, she works with clients to understand their business challenges and provide data-driven solutions using SAS Viya and open source tools. She also collaborates with internal teams to develop and deliver innovative use cases for the health and life sciences sector. In her prior role, she spearheaded the data science team at a global consultancy specializing in regulatory affairs and quality management. Here, she leveraged machine learning and text analytics to automate and optimize diverse processes and workflows, fostering a deep expertise in regulatory data science. Soundarya holds a master's degree in Informatics from the University of Freiburg, Germany, and has published a book chapter on IoT-based sensor systems for healthcare applications. She is passionate about leveraging data and analytics to improve health outcomes and enhance patient care.
Global Health & Life Sciences Practice
Focused on methods and application of unstructured data and text analytics, development of intelligent decision support systems for health care & life sciences by closely collaborating with SAS.
+ Developing a methodology to automate the labelling of various time series data and integrating the trained machine learning model with a user interface. + Training the networks using LSTM and CNN. +Working towards deploying the trained model in the production environment.
Title - Automated Extraction of Data Point Names using Machine learning methods. The objective is to classify various unevenly spaced time series signals obtained from different sensors and buildings in the field of Building automation systems. The concrete problem formulation would be to develop a methodology for automatically labeling the complex data sources based on raw time series data together in a supervised learning method. + Automatic building sensor classification using time series data. + Evaluated Random forest, SVM, KNN and MLP on the selected handcrafted features and analyzed their performance. + Investigated the strength of Convolutional Neural Networks (CNN) with temporal convolutions to learn features automatically. Achieved comparable performance with statistical models. + Achieved 98.6% accuracy on test data. + Proposed an entropy based approach to reduce misclassification rate iteratively in cases where ground truth is unknown. A new measure is introduced to assess the proposed approach and results showed this supervised learning approach is highly helpful to reduce the human labeling effort further.
Responsible for analysing the trends and patterns in the data for detecting oscillations in a time series signal using python.
LSB Modification in Steganography - The project explores a fraction of the art of steganography and carried out hiding and unhiding the secret text within the 24 bit color BMP image using LSB (least significant bit) algorithm. The aim was to avoid any change in the visual perception of the image even after manipulating the bits.