KANISHK SAXENA

Masters in Data Science graduate from the University of Glasgow with 5 years of experience as a Software Developer and Data Engineer at prestigious MNCs. Committed, self motivated and enthusiastic professional dedicated to contributing to team success through hard work, technical, analytical, and communication skills. Proficient in engaging with third-party vendors, stakeholders, and cross-functional teams to extract valuable insights. Committed to staying at the forefront of the dynamic field of data science, with a passion for continuous learning and professional growth.

• Hindi
• English

https://github.com/KanishkSaxena/data-analysis/tree/main

Predictive Modelling; Risk Assessment | Quadratic Weighted Kappa, Dense Neural Networks, Ensemble Methods:

• Aimed to categorize individuals into risk levels based on personal and health data using traditional machine learning models with neural networks and a multi-tier pipeline for stacking and ensemble. Implemented binary classification, multi-classification, and regression tasks using neural networks and traditional models. Utilized Quadratic Weighted Kappa metric for evaluation and optimized the model's performance by applying offset calculated with fmin Powell method.

Topic Modelling | Python, NLP, nltk, sklearn, SciPy, spaCy, Seaborn, pyALDavis

• Analyzed the data for short text topic modeling and performed long text topic modeling by clustering the same tweets using a single-pass approach. Generated topics for both datasets and conducted a comparative performance analysis between the two models.

Big Data analysis | Java, Apache Spark

• Developed a batch-based text search and filtering pipeline in Apache Spark, The core objective of this pipeline was to process large collection of text documents and a set of user-defined queries, resulting in the retrieval of the top 10 documents. This was achieved by implementing tokenization, utilizing a ranking model and resolving duplicates based on textual distance in document titles (using a provided comparison function). Achieved an outstanding performance milestone by processing 10 GB of data in just 50 seconds, demonstrating the pipeline's efficiency.

Deep learning Cancer Cell Classification | CNN, Ray parameter tuning, torchvision, Captum

• Trained and compared two deep neural networks which can take a 100x100 pixel images with a cell nuclei and classify wether tissue is normal or cancerous. One approach was to segment out the nuclei of cells from these images and classify them into different cell types. Comparison between two model where one was a small custom ConvNet and other was made from existing torchvision model which was pre-trained on ImageNet and further trained on the nucleus data – essentially using transfer learning.

Text processing techniques and its applications| NLP, HuggingFace, RoBerta, Transformers

• Text-as-Data project to build and evaluate different classifier's performance, like dummy classifier, svc, random forest and logistic regression, implemented k-means clustering also built a deep learning based approach BERT utilizing HuggingFace's 'feature-extraction' pipeline with the 'roberta_base' model to encode text documents for improved classification accuracy.