GPT-4o: Your Data Science Assistant

Introduction:

Data scientist constantly seeks new tools or methods to streamline exploratory data analysis process lead to gaining deeper insights into datasets. The rise of Generative AI in the last one year paved the way of bringing innovative ideas to aid data professionals by harnessing the power of Large Language Models(LLMs). From exploring a dataset to getting much needed coding assistance, the models have shown potential to evolve as an AI assistant for a data scientist who continuously thrive to enrich their data practices. In this article, I shared a glimpse of GPT-4o model's usability in performing quick exploratory check on a dataset, providing much needed insights to a data scientist for making judgment on the predictive modelling step.

Before I proceed further, let me give you a short description of the model. GPT-4o is OpenAI’s latest LLM. The 'o' in GPT-4o stands for "omni"—Latin for "every"—referring to the fact that this new model can accept prompts that are a mixture of text, audio, images, and video. This is a significant upgrade for its predecessor GPT-4 as the previous model interface used separate models for different content types. OpenAI unveiled their new flagship model on May 13th, 2024 and is presented as a free-to-use model that surpasses the capabilities of GPT-4 in many areas. However, users are recommended to get the paid subscription in order to avail the full model capability.

Dataset used:

For this exercise, I used the famous Boston House Price Dataset which involves the prediction of a house price in thousands of dollars given details of the house and its neighborhood. Let's scroll further to find out what GPT-4o model is going tell you about the dataset via actionable prompts.

Initial Data Exploration:

Data science work begins with exploring the dataset, enabling the researcher to gain understanding of the overall structure, column types and descriptions. The prompt used to fetch the information:

This is Boston housing price dataset. Please provide an initial data exploration summary derived from this dataset.

GPT-4o model provided an overview of the dataset comprised of the number or observations, features and a brief description of each feature.

Note: Evidently the dataset used in this exercise is very popular and readily available on the internet thus the GPT model was already pre-trained with the model description before.

Furthermore, the same prompt produced the following output, which not only eliminates the additional step of exploring persistent missing values in the dataset but also provides guidance on determining the next steps in the data science workflow.

Correlation matrix and outliers detection:

Understanding the correlations among features in a dataset is crucial before moving to the modeling step. Additionally, data scientists must check for prevalent outliers to ensure that data sparsity or spread will not impact model performance in later stages. Therefore, the next prompt, as mentioned below, sheds light on this key step:

Please provide insights on the correlation matrix and potential outliers that possibly underpin performance of a predictive model that uses this dataset as an input.

GPT-4o output, as demonstrated below, manifests high and low correlation between certain features, enabling a data scientist to consider those variables to opt in or opt out during the model development phase.

Outlier detection by the GPT-4o model:

Furthermore, GPT-4o won't be able to render plots on the user interface to visualize correlation matrix or outlier detection. However, it can provide coding template that can be repurposed to develop data visualization reflecting on the steps discussed above. Sample screenshot:

Note: It is advisable to use any code generated by ChatGPT only after performing a thorough review and applying your own professional judgment.

Feature engineering and model predictors:

Feature engineering step is important to ensure the highly correlated features are isolated from the model prediction as the reduction of multicollinearity would help with alleviating model skewness issue. GPT-4o model suggests the following steps and variables to ensure feature selection step benefits data scientist to pick features with strong relationships with the target variable (medv).

Predictive model recommendation:

Here comes the final step, where a data scientist can seek a model recommendation from GPT-4. Boston housing price prediction is a classic case of a linear regression model exercise. However, a data scientist still needs to design a blueprint of the model development steps to ensure the model is developed with the right predictors, followed by reasonable model evaluation steps. Prompt below:

Please provide recommendation on what type of predictive model should be used to determine the housing price dataset supported by statistical analysis.

The following output from GPT-4 is straightforward for a data scientist to interpret. It outlines preliminary steps for selecting a linear regression model, provides tips for addressing multicollinearity caused by strongly correlated variables in the dataset, suggests non-linear models, and includes basic model evaluation techniques applicable to supervised learning.

Conclusion:

It is quite evident that the GPT-4 model could potentially become an essential tool in the data science toolkit in the near future. However, we should not overlook the risks related to data security and privacy before utilizing this model to analyze company proprietary datasets. For the time being, data scientists can use this platform to analyze public datasets if they align with their model development work. Considering the data privacy concerns, data scientists should not be encouraged to upload private datasets to the GPT interface without obtaining approval from their respective organizations.

Disclaimer:

The AI-generated data science recommendations provided here do not constitute professional advice and should not be considered as a substitute for consulting with qualified experts in the field. Always seek appropriate approvals from your organization before using AI tools for sensitive data analysis.