Data Modeling And Visualization

Problem

Many data tasks stall at the ambiguous part: after receiving a CSV or Excel file, an AI may jump directly into modeling, plotting, or conclusions without defining the target variable, checking data quality, or confirming the analytical goal. This can lead to regression on categorical data, ignored missing values, causal claims from correlation, or charts without business interpretation.

How It Works

The skill constrains the workflow to a standard data-science path: data understanding → EDA → model selection → model validation → visualization and explanation.

• Boundary handling first: check rows, column names, and missing values; pause when missing data is high, and switch to frequency tables, cross-tabs, or chi-square tests for fully categorical data.
• EDA before modeling: use distributions, correlations, heatmaps, boxplots, and cross-tabs, focused on the user-specified target variable.
• Explainable model selection: choose linear regression, random forests, logistic regression, trend decomposition, or PCA based on linearity, classification, time series, or exploratory needs, with stated reasoning.
• Validation and reporting: report metrics such as R², RMSE, MAE, accuracy, and F1, plus key findings, business interpretation, and limitations.

Boundaries

It is suited to exploratory analysis and basic modeling on small-to-medium tabular data. It is not intended to replace XGBoost, neural networks, large-scale distributed processing, or specialized geographic/network visualization; those cases require extended skills.