Shivani Pathak

Data science and artificial intelligence professional with expertise in machine learning, natural language processing, and data analysis. Former assistant professor with a proven track record in teaching AI and machine learning concepts while leading technical projects. Strong capabilities in technical reasoning, comparative analysis, and data-driven decision-making. Recently completed a professional certificate in data science and analytics, actively pursuing opportunities to apply skills in solving real-world challenges.

• AI Based Payment Fraud Detection using Machine Learning: Developed a fraud detection model using a dataset of about 1 million financial transactions. Performed data preprocessing and handled class imbalance using SMOTE to improve model performance. Compared multiple machine learning models including Logistic Regression, Random Forest, SVM, Decision Tree, and XGBoost. Evaluated models using accuracy, precision, recall, and AUC, where XGBoost achieved the best performance.
• Patient Appointment Overbooking Optimization | Python, SciPy: Developed a statistical model in Python to analyze patient no-show rates using a binomial distribution. Applied SciPy functions (PMF, CDF, PPF) to determine the optimal number of daily bookings for a physician with a 30-patient capacity while limiting overbooking risk to 5%. Visualized probability distributions using Matplotlib to support scheduling decisions and improve operational efficiency.
• Automobile Customer Preference Analysis – Python: Analyzed customer survey data using Python to identify patterns in automobile feature preferences. Applied clustering techniques including K-Means and hierarchical clustering, used PCA for dimensionality reduction, and created visualizations such as heatmaps, dendrograms, and scatter plots to interpret relationships among vehicle attributes.
• Healthcare Data Analysis: Diabetes Prediction Model: Built a predictive model using multiple linear regression to estimate diabetes disease progression from clinical and demographic variables. Used Python libraries including Pandas, NumPy, Statsmodels, and Scikit-Learn to preprocess data, train regression models, perform statistical significance testing, and evaluate performance using R² metrics.
• Loan Default Prediction Model – Python: Built a logistic regression model to predict loan defaults using FICO scores. Conducted data analysis, trained the model on a split dataset, and evaluated performance using confusion matrix and accuracy metrics. Adjusted probability thresholds to improve classification outcomes for lending decisions.
• Music Recommendation System – Collaborative Filtering (Python): Built a recommendation system using collaborative filtering to predict missing user ratings for songs. Applied matrix factorization techniques with SoftImpute, performed data preprocessing and normalization, and evaluated model performance using R² and MSE on training and validation datasets.
• Diabetes Classification Model – XGBoost (Python): Built an XGBoost classification model to predict diabetes using medical diagnostic data. Performed data preprocessing, trained the model with early stopping, and evaluated performance using accuracy and ROC-AUC metrics. Visualized feature importance to interpret model predictions.
• ECG Abnormality Detection Using Machine Learning – Python: Built a binary classification model to predict abnormal ECG signals from structured clinical data. Conducted data preprocessing, including class distribution analysis and train-test splitting, and trained a machine learning model using Scikit-Learn. Evaluated performance using accuracy, confusion matrix, and classification metrics, and interpreted results to assess model reliability in detecting abnormal heart conditions.
• Deep Learning Image Classifier – TensorFlow/Keras: Built a multi-class image classification model using a fully connected neural network to classify 28×28 grayscale images. Implemented data normalization, label encoding, and model training using TensorFlow/Keras. Designed a neural network architecture with dense layers, ReLU activation, and softmax output, and optimized model performance using cross-entropy loss and gradient-based optimization.
• NLP-Based Sentiment Classification Model – TensorFlow/Keras: Built a deep learning model for sentiment analysis using IMDb movie reviews. Applied text vectorization with multi-hot encoding, constructed a feedforward neural network with ReLU activation and dropout regularization, and trained the model using mini-batch gradient descent. Tuned model parameters such as hidden layers, epochs, and feature representation (unigrams/bigrams) to improve classification accuracy.
• Intent Classification & Slot Filling Using Transformers: Built a transformer encoder model to classify user queries and predict slot entities on the ATIS dataset. Applied attention mechanisms and positional embeddings to improve accuracy on unseen inputs, achieving high overall and slot-level accuracy.
• Parser Evaluation Study, Comparative analysis of statistical and rule based parsers to identify syntactic patterns and improve language processing frameworks.
• Word Alignment in Multilingual Corpora, English Hindi and Punjabi Hindi word alignment and NLP resource development.
• Improving Fluency of Machine Translation, Applied statistical machine translation methods and language modeling techniques.
• Steganography and Watermarking System, Implemented secure data hiding techniques using Java.

• Hindi
• English