Introduction

Overview of this analysis is to gain insight on climate change trends over time in the United States.

• Can we quantify the human impact on climate change in the USA?
• Does human impact affect how states individually experience climate change?
• Are there climate change trends evident in temperature over time?
• Is there a possibility to predict any other feature like CO2, year, state disasters from the datasets?

Climate Trends

Temperature Findings

Examine temperature of the States over time using the slide bar.

Emissions Findings

Press play to see the change of CO2 emissions and changes in Renewable and Primary energy use over time.

Models

The datasets used for this analysis are pertaining to climate change in the United States of America. Human interventions contributing to climate change in the contiguous USA are specifically considered along with time, location, gdp and population. After the data was processed, features were selected for thorough analysis based on the understanding of the features and its likely contribution to climate change. Given that this dataset has lot to be explored, to gain understanding Unsupervised Machine learning model was employed. Post Unsupervised ML, it was determined that temperature, CO2, year, state, disasters of a state could be considered as potential targets for supervised ML with rest of the features supplied as features.

Predictions

After conducting an in-depth analysis of the features and testing various machine learning model combinations, it was learned that predicted temperature with just state and year has a near perfect R-squared of 97.55. This led to the inference that the dataset has confounding features and some of the features of the dataset are behaving as proxies for US states(location). Statistically, this causes spurious association. Considering the above observations, a model was built to predict temperature based on only the human impact features CO2 emissions, population, real GDP, renewable and non-renewable energy consumption features. The R-Squared for this model was found to be 43.68%. Human-related features were able to predict the state name with 98.48% accuracy using just a simple neural network consisting of a single hidden layer of 100 nodes. This implies that state name (location) has a great influence on GDP, population, CO2 emission and energy consumption. It is reasonable to believe that location contains unique features like demographics and information about the local economy which influence these human-related features.

Energy Consumption

Compare average renewable and non-renewable energy use between classes by selecting a class or classes using the filter below.

State Comparison

Compare states based on renewable and non-renewable energy given their average temperature and CO2 emissions selecting them on the map below.

Energy By Capita

Compare states based on renewable and non-renewable energy per capita.