Forecasting with Random Forests

When it comes to forecasting data (time series or other types of series), people look to things like basic regression, ARIMA, ARMA, GARCH, or even Prophet but don’t discount the use of Random Forests for forecasting data.

Random Forests are generally considered a classification technique but regression is definitely something that Random Forests can handle.

For this post, I am going to use a dataset found here called Sales Prices of Houses in the City of Windsor (CSV here, description here).  For the purposes of this post, I’ll only use the price and lotsize columns. Note: In a future post, I’m planning to resist this data and perform multivariate regression with Random Forests.

To get started, let’s import all the necessary libraries to get started. As always, you can grab a jupyter notebook to run through this analysis yourself here.

Now, lets load the data:

Again, we are only using two columns from the data set – price and lotsize. Let’s plot this data to take a look at it visually to see if it makes sense to use lotsize as a predictor of price.

Housing Data Visualization

Looking at the data, it looks like a decent guess to think lotsize might forecast price.

Now, lets set up our dataset to get our training and testing data ready.

In the above, we set X and y for the random forest regressor and then set our training and test data. For training data, we are going to take the first 400 data points to train the random forest and then test it on the last 146 data points.

Now, let’s run our random forest regression model.  First, we need to import the Random Forest Regressor from sklearn:

And now….let’s run our Random Forest Regression and see what we get.

Let’s visualize the price and the predicted_price.

price vs predicted price

That’s really not a bad outcome for a wild guess that lotsize predicts price. Visually, it looks pretty good (although there are definitely errors).

Let’s look at the base level error. First, a quick plot of the ‘difference’ between the two.

Price vs Predicted Difference

There are some very large errors in there.  Let’s look at some values like R-Squared and Mean Squared Error. First, lets import the appropriate functions from sklearn.

Now, lets look at R-Squared:

R-Squared is 0.6976…or basically 0.7.  That’s not great but not terribly bad either for a random guess. A value of 0.7 (or 70%) tells you that roughly 70% of the variation of the ‘signal’ is explained by the variable used as a predictor.  That’s really not bad in the grand scheme of things.

I could go on with other calculations for error but the point of this post isn’t to show ‘accuracy’ but to show ‘process’ on how how to use Random Forest for forecasting.

Looks for more posts on using random forests for forecasting.


If you want a very good deep-dive into using Random Forest and other statistical methods for prediction, take a look at The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition (Amazon Affiliate link)

Quick Tip: SQLAlchemy for MySQL and Pandas

SQLAlchemy LogoFor years I’ve used the mysql-python library for connecting to mysql databases.  It’s worked well for me over the years but there are times when you need speed and/or better connection management that what you get with mysql-python.  That’s where SQLAlchemy comes in.

Before diving into this, if you are doing things that aren’t dependent on speed (e.g., it doesn’t matter if it takes 1 second to connect to the database and grab your data and close the database) then you can easily ignore this tip. That said, if you have multiple connections, that connect time can add up.

For example, I recently had an issue where it was taking 4.5+ seconds to connect to a database, run analysis and spit out the results. That’s not terrible if its something for you only but if its a production system and speed is a requirement, that might be too long (and it IS too long).

When I did some analysis using python’s timer() I found that more than 50% of that 4.5 seconds time was in establishing database connections so I grabbed my trusty SQLAlchemy toolkit and went to work.

For those of you that don’t know, SQLAlchemy is a ‘python SQL toolkit and Object Relational Mapper’ (ORM) that is supposed to make things easier when working with SQL databases. For me, the ORM aspect tends to make things more difficult so I tend to stick with plain SQL queries but the SQL toolkit aspect of SQLAlchemy makes a lot of sense and add some time savings when connecting to a SQL database.

Before we get into the SQLAlchemy aspects, let’s take a second to look at how to connect to a SQL database with the mysql-python connector (or at least take a look at how I do it).

First, let’s setup our import statements. For this, we will import MySQLdb, pandas and pandas.io.sql in order to read SQL data directly into a pandas dataframe.

Next, let’s create a database connection, create a query, execute that query and close that database.

This is a fairly standard approach to reading data into a pandas dataframe from mysql using mysql-python.  This approach is what I had been using before when I was getting 4.5+ seconds as discussed above. Note – there were multiple database calls and some analysis included in that 4.5+ seconds. A basic database call like the above ran in approximately 0.45 seconds in my code that I was trying to improve performance on and establishing the database connection was the majority of that time.

 To improve performance – especially if you will have multiple calls to multiple tables, you can use SQLAlchemy with pandas.   You’ll need to pip install sqlalchemy if you don’t have it installed already. Now, let’s setup our imports:

Now you can setup your connection string to your database for SQLAlchemy, you’d put everything together like the following:

where USER is your username, PW is your password, DBHOST is the database host and  DB is the database you want to connect to.

To setup the persistent connection, you do the following:

Now, you have a connection to your database and you’re ready to go. No need to worry about cursors or opening/closing database connections. SQLAlchemy keeps the connection management aspects in for you.

Now all you need to do is focus on your SQL queries and loading the results into a pandas dataframe.

That’s all it takes.  AND…it’s faster.  In the example above, my database setup / connection / query / closing times dropped from 0.45 seconds to 0.15 seconds.  Times will vary based on what data you are querying and where the database is of course but in this case, all things were the same except for mysql-python being replaced with SQLAlchemy and using the new(ish) read_sql_query function in pandas.

Using this approach, the 4.5+ seconds it took to grab data, analyze the data and return the data was reduced to about 1.5 seconds. Impressive gains for just switching out the connection/management method.

Local Interpretable Model-agnostic Explanations – LIME in Python

When working with classification and/or regression techniques, its always good to have the ability to ‘explain’ what your model is doing. Using Local Interpretable Model-agnostic Explanations (LIME), you now have the ability to quickly provide visual explanations of your model(s).

Its quite easy to throw numbers or content into an algorithm and get a result that looks good. We can test for accuracy and feel confident that the classifier and/or model is ‘good’…but can we describe what the model is actually doing to other users? A good data scientist spends some of their time making sure they have reasonable explanations for what the model is doing and why the results are what they are.

There’s always been a focus on ‘trust’ in any type of modeling methodology but with machine learning and deep learning, many people feel like the black-box approach taken with these methods isn’t as trustworthy as other methods.  This topic was addressed in a paper titled Why Should I Trust You?”: Explaining the Predictions of Any Classifier, which proposes the concept of Local Interpretable Model-agnostic Explanations (LIME). According to the paper, LIME is ‘an algorithm that can explain the predictions of any classifier or regressor in a faithful way, by approximating it locally with an interpretable model.’

I’ve used the LIME approach a few times in recent projects and really like the idea. It breaks down the modeling / classification techniques and output into a form that can be easily described to non-technical people.  That said, LIME isn’t a replacement for doing your job as a data scientist, but it is another tool to add to your toolbox.

To implement LIME in python, I use this LIME library written / released by one of the authors the above paper.

I thought it might be good to provide a quick run-through of how to use this library. For this post, I’m going to mimic “Using lime for regression” notebook the authors provide, but I’m going to provide a little more explanation.

The full notebook is available in my repo here.

Getting started with Local Interpretable Model-agnostic Explanations (LIME)

Before you get started, you’ll need to install Lime.

Next, let’s import our required libraries.

Let’s load the sklearn dataset called ‘boston’. This data is a dataset that contains house prices that is often used for machine learning regression examples.

Before we do much else, let’s take a look at the description of the dataset to get familiar with it.  You can do this by running the following command:

The output is:

Now that we have our data loaded, we want to build a regression model to forecast boston housing prices. We’ll use random forest for this to follow the example by the authors.

First, we’ll set up the RF Model and then create our training and test data using the train_test_split module from sklearn. Then, we’ll fit the data.

Now that we have a Random Forest Regressor trained, we can check some of the accuracy measures.

Tbe MSError is: 10.45. Now, let’s look at the MSError when predicting the mean.

From this, we get 80.09.

Without really knowing the dataset, its hard to say whether they are good or bad.  Since we are really most interested in looking at the LIME approach, we’ll move along and assume these are decent errors.

To implement LIME, we need to get the categorical features from our data and then build an ‘explainer’. This is done with the following commands:

and the explainer:

Now, we can grab one of our test values and check out our prediction(s). Here, we’ll grab the 100th test value and check the prediction and see what the explainer has to say about it.

LIME Explainer for regression
LIME Explainer for regression

So…what does this tell us?

It tells us that the 100th test value’s prediction is 21.16 with the “RAD=24” value providing the most positive valuation and the other features providing negative valuation in the prediction.

For regression, this isn’t quite as interesting (although it is useful). The LIME approach shows much more benefit (at least to me) when performing classification.

As an example, if you are trying to classify plants as edible or poisonous, LIME’s explanation is much more useful. Here’s an example from the authors.

LIME explanation of edible vs poisonous
LIME explanation of edible vs poisonous

Take a look at LIME when you have some time. Its a good library to add to your toolkit, especially if you are doing a lot of classification work. It makes it much easier to ‘explain’ what the model is doing.