The main task in this challenge is to use machine learning to enhanced existing trading signals in an automated trading algorithm.
The relavent python notebooks is titled machine_learning_trading_bot.ipynb
This code is to be run on Python 3.7.13
The following Python Libraries were also imported and used:
import pandas as pd
from path import Path
import numpy as np
import hvplot.pandas
import matplotlib.pyplot as plt
from sklearn import svm
from sklearn.preprocessing import StandardScaler
from pandas.tseries.offsets import DateOffset
from sklearn.metrics import classification_report
- Establish a Baseline Performance of the algorithm
- Tune the Baseline Trading Algorithm
- Evaluate a New Machine Learning Classifier
The initial classifier model used was SVC from sklearn, which is a type of support vector machine learning method, which calculates the best hyperplane to seperate data into different classes. By evaluating this model's precsion and recall, we found that is performed worse at predicting whent to short that it did at predicting when to buy (whihc itself was only slightly above 50%). The recall was also only 4% for shorting.
The second model we created was created using Logistical Regression.
Based on the classification report, the Logistic Regression model is only slightly better than 50% accurate. But it is better than the intial SVM, and as we can see in the notebook charts, the Logistic Regression model provides a return that is comperable to the actual returns, and resulted in a more consistant return from 2016 through 2020.