Bitcoin Price Prediction

Project 01

  • Developed predictive models (ridge regression, lasso regression, random forest regression, decision tree regression) to forecast Bitcoin prices based on macroeconomic factors.
  • DBSCAN clustering algorithm to identify and remove anomalous data points.
  • Performed data cleaning and preprocessing on six economic datasets, aligning over 8,000 data points.
  • PCA analysis on the dataset resulting in a reduction of features by 25% without compromising model accuracy.

    • Project Report: View Report

    OPTIMIZING WAREHOUSE OPERATIONS FOR DELTA TECHOPS

    Project 02

  • Developed a Tableau dashboard, integrated with the client’s SAP database, to track 500+ daily incoming parts improving decision-making efficiency by 40% through real-time insights into part types, quantities, and delivery schedules.
  • Directed data preprocessing and manipulation across 1TB of warehouse operational data to reveal critical operational bottlenecks.
  • Developed a simulation model to determine optimal worker allocation across all processing lines leading to a 25% increase in efficiency.
  • Overall, these deliverables resulted in an annual saving of approximately $450,000 for Delta TechOps.

    • CREDIT SCORING MODEL FOR SMALL BUSINESS LOAN APPROVAL

    Project 03

  • Developed a logistic regression model in R to predict loan default probability among small businesses.
  • Conducted data cleaning, handling missing fields and null values on over 50,000 data points.
  • Performed EDA using a correlation matrix and scatter plots, revealing correlations between key predictors.
  • Experimented with various link functions, and variable transformations to find the most effective model.
  • Applied k-fold cross-validation methods to ensure model robustness and reliability.

    • Project Report: View Report

    PREDICTING REGIONAL SONG POPULARITY WITH SPOTIFY API

    Project 04

  • Developed a Random Forest model to predict song popularity by region, analyzing over 220,000 songs using data from the Spotify API.
  • Employed D3 to create an interactive world map visualization, enabling dynamic visualization of song popularity trends across different global regions.
  • Conducted A/B testing to optimize data visualization, choosing a Cleveland dot plot for its clarity
  • Implemented a data processing pipeline, involving data cleaning, NaN value handling, and feature encoding using OneHotEncoder to prepare the dataset.

    • Project GitHub: View GitHub

    TEXT-TO-SPEECH SYNTHESIS WITH NEURAL NETWORK FUSION

    Project 05

  • Developed an advanced text-to-speech synthesis model by combining key elements of two existing neural network architectures to enhance speech clarity and naturalness.
  • Employed the LJSpeech dataset for model training and testing, consisting of 13,100 short audio clips from a single-speaker.
  • Conducted rigorous performance evaluation using Mel Cepstral Distortion (MCD) metric, achieving an average MCD of 4.9777.

    • Project Report: View Report