AIN311 Machine learning in Sustainability project Week 3: Experiments with Linear Regressor and Random Forest Regressor

Welcome back, fellow sustainability enthusiasts! It’s Week 3 of our Machine Learning in Sustainability project and things are heating up (not just the climate)! This week we are heading into the world of Linear Regression and Random Forest Regression. Buckle up for our discoveries!

🌱 Linear Regressor: Dancing with Data

Our first exploration led us to the world of Linear Regression, a trusty companion in predicting numerical values. Tasked with unraveling the mysteries of sustainability metrics, our Linear Regressor embarked on a journey of discovery.

We examined the problem under 4 Settings which are: All columns used, only most correlated column, 2 most correlated columns and Polynomial Regression for most correlated column.How did those settings influence the predictive power of our model? The answers lie within the experiment results.

🌳 Random Forest Regressor: A Forest of Possibilities

As the Linear Regressor took center stage, we introduce our second protagonist, the Random Forest Regressor. This ensemble method creates a symphony of decision trees, perfect for predicting both classification and regression outcomes.

We examined our second protagonist under 4 Settings which are: All columns, only most correlated column, 2 most correlated columns and Hyperparameter tuned all columns. We had a problem with hyperparameter tuning’s computing time. We will carry out further experiments with it. We will analyze the results under results showdown!

☀️Results Showdown

It’s our time to shine! Did the Linear Regressor bring clarity to sustainability metrics? Was the Random Forest Regressor the guardian of the environmental realm?

Our first participant is Linear regression which had the following results:

We can observe that using all columns has incredible impact on lowering our error rate. Thus giving the best results. We think that we will be using all the columns for our model from now on but that is yet to be changed with other results.

Here comes our second and final participant for this week ! Random Forest Regression which had the following results:

We can observe that using all columns has some impact on lowering our error rate but it is still too much compared to Linear regression. Only question mark we have is If it will benefit us to change Hyperparameter grid for good. For now Linear regression seems like a flawless winner both in computational time and error rates.

🌐 Looking Ahead: Neural Nets on the Horizon

As we bid farewell to Week 3 we look towards the future. Week 4 promises an encounter with a 1DCNN model we will see what can be done.

Stay curious, stay sustainable, and stay tuned for our expeirments.