BTC Time-Series Forecasting
Developed a comprehensive time-series forecasting application for Bitcoin price evaluation. The project ingests historical data and dynamically splits it to compare predictive capabilities between statistical autoregressive models (ARIMA) and non-linear machine learning approaches (Decision Tree Regressor). The application features a TradingView-styled interactive Streamlit dashboard that visualizes high-definition forecasts, separating true historical sequences, backtest accuracy, and future runs with uncertainty zones.
Technologies Used
Problem Statement
Cryptocurrency analysts often struggle with fragmented data pipelines and rigid forecasting tools that lack intuitive visual backtesting. Existing solutions either require extensive manual formatting or fail to visually distinguish between training data, backtest predictions, and out-of-sample future forecasts effectively.
Solution
This project provides an end-to-end forecasting pipeline wrapped in a premium interactive web application. It automatically ingests Kaggle-style CSV data, splits environments, and runs competitive predictions between ARIMA and Decision Trees. The Plotly-powered dashboard overlays future forecasts and confidence intervals directly onto historical data, providing traders and analysts with immediate, visually-striking insights into model reliability and future trends.
Key Features
Automated CSV data ingestion and chronologically sorted validation
ARIMA and Decision Tree Regressor competitive modeling
TradingView-style interactive Plotly dashboard with hover details
Dynamic train-test splitting and variable forecast horizons (7-90 days)
Engineering Challenges
Handling non-stationary crypto data requiring careful differencing (ARIMA order tuning)
Rendering high-density interactive Plotly charts smoothly within Streamlit
Calculating and shading accurate confidence intervals for future horizons
Results & Metrics
Delivered a robust, low-latency predictive dashboard for crypto analysis
Successfully established a reproducible backtesting environment
Enabled non-technical users to run complex time-series models interactively
Lessons Learned
Statistical models (ARIMA) often capture linear trends better than simple tree-based regressors in time-series
Interactive data visualization is as critical as model accuracy for financial tooling
Streamlit combined with Plotly provides a powerful rapid-prototyping stack for ML apps