Skip to content

🚀 Projects

  • Objective: Built a classification model to predict telecom customer churn using a real-world dataset.
  • Approach:
    • Cleaned and preprocessed data: handled missing values, label-encoded categorical features, normalized numerical columns.
    • Trained multiple models (Logistic Regression, Decision Tree, Random Forest).
    • Selected the best-performing model using cross-validated F1-score.
  • Metrics Used: Accuracy, Precision, Recall, F1-score, Confusion Matrix.
  • Libraries/Tools: Python, Pandas, Scikit-learn
  • Dataset: ~7,000 rows
  • 🎯 Delivered actionable insights such as how contract type and monthly charges affect churn.
  • Objective: Predicted international education costs using regression modeling to uncover pricing dynamics.
  • Approach:
    • Performed EDA to identify key factors (e.g., country, degree type, institution type).
    • Engineered features: imputation, encoding, scaling.
    • Compared ANN model with Linear Regression and Random Forest.
  • Metrics Used: MAE, RMSE
  • Libraries/Tools: Python, Pandas, Scikit-learn, Seaborn, Plotly
  • Dataset: ~1,000 rows
  • 🧠 Achieved ~90% test accuracy using ANN.
  • Objective: Explored hiring trends, regional insights, and in-demand skills from LinkedIn job postings.
  • Approach:
    • Cleaned inconsistent job titles using RegEx.
    • Performed grouped analysis to extract trends and demand.
    • Summarized technical/soft skills from descriptions.
  • Visuals: Bar plots, heatmaps, distributions.
  • Libraries/Tools: Python, Pandas, Seaborn, Plotly, RegEx
  • Dataset: ~300 rows
  • 🔍 Offered a data-driven view of the data science job market.
  • Objective: Developed a binary text classification model to detect fake news using semantic embeddings.
  • Approach:
    • Preprocessed text (tokenization, stop-word removal, lemmatization).
    • Generated 300-dim Word2Vec embeddings from Google News corpus.
    • Built document-level vectors by averaging word embeddings.
    • Trained and tuned SVM and Logistic Regression models.
  • Metrics Used: Accuracy, Precision, Recall, F1-score
  • Libraries/Tools: Python, Gensim, Scikit-learn, NLP
  • Dataset: ~17,000 rows
  • 🤖 Outperformed traditional bag-of-words classifiers.