bitcoin_image

Date: October 13, 2021

Table of Contents

  1. Project Planning
    1. Project Objectives
    2. Business Goals
    3. Audience
    4. Deliverables
  2. Executive Summary
  3. Acquire Data
    1. Data Dictonary
    2. Acquire Takeaways
  4. Prepare Data
    1. Prepare Takeaways
  5. Data Exploration
    1. Correlations
    2. Pairplot
    3. Explore Takeaways
  6. Modeling
    1. Last Observed Value
    2. Rolling & Moving Average
    3. Holt’s Linear Trend
    4. Previous Cycle
  7. Delivery
    1. Root Mean Square Deviation
    2. Conclusions & Next Steps
    3. Replication

Project Planning

✓ 🟢 Plan ➜ ☐ Acquire ➜ ☐ Prepare ➜ ☐ Explore ➜ ☐ Model ➜ ☐ Deliver

Project Objectives

  • For this project we will be working with historical price and volume data from Bitcoin between 01-01-2012 & 03-31-2021, these are Bitstamp prices and all are annotated in USD.
  • The primary focus is to see if Bitcoin price can be predicted with any reliability or if there is any cyclical observations within Bitcoin pricing or volume.
  • The csv data can be downloaded from Kaggle here.

Business Goals

  • Create models that are better at predicting Bitcoin price than the baseline.
  • Put these models into a Juypter notebook and make the project replicable.

Audience

  • Data science professionals as well as any curious cat.

Deliverables

  • A clearly named final notebook. This notebook will be what you present and should contain plenty of markdown documentation and cleaned up code.
  • A README that explains what the project is, how to reproduce you work, and your notes from project planning.
  • A Python module or modules that automate the data acquisistion and preparation process. These modules should be imported and used in your final notebook.

Executive Summary

Goals

  • This project is to utilize machine learning modeling to predict the avgerage price of Bitcoin, better than the baseline.
  • Abstract the functions to sub python scripts to have a clean presentation, and throughly document.

Findings

  • The dataset had quite a few null values, but utilizing a forward fill method offered an easy and effective remedy. Otherwise, this dataset was quite clean.
  • Another key aspect of predicting Bitcoin’s price was the percent change within a time interval.
  • The best performing model was Holt’s Linear Trend that was fine tuned for the dataset via slope and level smoothness, with an RMSE of 53.

Acquire Data

✓ Plan ➜ 🟢 Acquire ➜ ☐ Prepare ➜ ☐ Explore ➜ ☐ Model ➜ ☐ Deliver

Data Tail

Timestamp Open High Low Close Volume_(BTC) Volume_(Currency) Weighted_Price
2021-03-30 23:56:00 58714.3 58714.3 58686 58686 1.38449 81259.4 58692.8
2021-03-30 23:57:00 58684 58693.4 58684 58685.8 7.29485 428158 58693.2
2021-03-30 23:58:00 58693.4 58723.8 58693.4 58723.8 1.70568 100117 58696.2
2021-03-30 23:59:00 58742.2 58770.4 58742.2 58760.6 0.720415 42333 58761.9
2021-03-31 00:00:00 58767.8 58778.2 58756 58778.2 2.71283 159418 58764.3

Data Dictonary

Feature Datatype Definition
Timestamp 4857377 non-null: datetime64[ns] start tiem of time window (60s window), in Unix Time
Open 3613769 non-null: float64 Open price at start time window
High 3613769 non-null: float64 High price within the time window
Low 3613769 non-null: float64 Low price within the time window
Close 3613769 non-null: float64 Close price at the end of the time window
Volume_(BTC) 3613769 non-null: float64 Volume of BTC transacted in this window
Volume_(Currency) 3613769 non-null: float64 Volume of corresponding currency transacted in this window
Weighted_Price 3613769 non-null: float64 VWAP - Volume Weighted Average Price

Takeaways from Acquire:

  • Target variable: avg_price
  • This dataframe currenly has 4,857,377 rows and 8 columns
  • There are 1,243,608 missing values.
  • All columns are float64 types of data.

Prepare Data

✓ Plan ➜ ✓ Acquire ➜ 🟢 Prepare ➜ ☐ Explore ➜ ☐ Model ➜ ☐ Deliver

  • Add additional columns of month, day_of_week, price_diff, price_delta, percent_change and day_num.
  • Filling the null values with the most recent value will likely be the best course of action.

New Data Dictionary

Feature Datatype Definition
Open 4857377 non-null: float64 Open price at start time window
High 4857377 non-null: float64 High price within the time window
Low 4857377 non-null: float64 Low price within the time window
Close 4857377 non-null: float64 Close price at the end of the time window
Volume_(BTC) 4857377 non-null: float64 Volume of BTC transacted in this window
Volume_(Currency) 4857377 non-null: float64 Volume of corresponding currency transacted in this window
Weighted_Price 4857377 non-null: float64 VWAP - Volume Weighted Average Price
day_of_week 4857377 non-null: object Verbose name of the week
day_of_week_num 4857377 non-null: int64 number representing the day of the week
month 4857377 non-null: object Month number and month name
month_num 4857377 non-null: int64 number representing the month of the year
price_diff 4857377 non-null: float64 Delta between the Close and Open (Close - Open)
price_delta 4857377 non-null: float64 Delta between the High and Low (High - Low)
day_num 4857377 non-null: int64 The numeric number of the day of the month
avg_price 4857377 non-null: float64 Avg price for the time period ([Open + Close] / 2)
percent_change 4857377 non-null: float64 Price difference / Open price represented as a percentage (price_diff / Open)

Prepare Takeaways

  • The data is now prepared to be input into the explore aspects of the data pipeline to evaluate what features we should use to potentially run time-series analysis on.

Explore Data

✓ Plan ➜ ✓ Acquire ➜ ✓ Prepare ➜ 🟢 Explore ➜ ☐ Model ➜ ☐ Deliver

Bitcoin Price over Time

Bitcoin_price

Hour vs Percent Change on a monthly basis

hour_vs_percent_change_monthly

Correlations

heatmap

Correlations of Average Price

Column Name avg_price
price_diff 0.0061726
percent_change 0.00543991
Volume_(Currency) -0.0475418
Volume_(BTC) -0.067203
price_delta -0.142155
day_num -0.346819
month_num nan

Pair Plot

pairplot

Explore Takeaways

  • It is apparent that the most correlated features with avg_price are day_num, price_delta, Volume_(BTC).
  • It also appears that the avg_price tends to trend upward during Oct, and downwards in Sept.

Modeling

✓ Plan ➜ ✓ Acquire ➜ ✓ Prepare ➜ ✓ Explore ➜ 🟢 Model ➜ ☐ Deliver

Last Observed Value

Average Price

last_observed_value

Percent Change

last_observed_value

Rolling/Moving Average

Average Price

rolling_value

Percent Change

rolling_value

Holt’s Linear Trend

Average Price

holt_value

Percent Change

holt_value_pct

Previous Cycle (6 months)

Average Price

previous_cycle

Percent Change

previous_cycle_pct

Project Delivery

✓ Plan ➜ ✓ Acquire ➜ ✓ Prepare ➜ ✓ Explore ➜ ✓ Model ➜ 🟢 Deliver

Root Mean Square Deviation Results

Average Price

rmse_results

Percent Change

rmse_pct_results

Conclusions & Next Steps

  • I found that there were large variations in certain models RMSE but the models that could be tuned (such as Holt’s and Previous Cycles) performed best after optimization.
  • If I had more time to work on this project, I’d continue doing more performance tuning and possibly aggregating other external data factors that could provide a higher level of correlation or model accuracy.

Replication

  • Download the the csv data from Kaggle here.
  • Download the acquire.py & prepare.py files and then run the final notebook.