Overview

Analysis Overview

LASSO Regression Configuration

Analysis overview and configuration

Configuration

Analysis TypeLasso
CompanyTest Company
ObjectiveIdentify which advertising channels drive sales using LASSO variable selection
Analysis Date2026-03-15
Processing Idlasso_test_20260315_114606
Total Observations300

Module Parameters

ParameterValue_row
alpha1alpha
n_folds10n_folds
lambda_choicelambda.1selambda_choice
standardizeTRUEstandardize
Lasso analysis for Test Company

Interpretation

Purpose

This LASSO regression analysis identifies which advertising channels drive sales by applying automatic variable selection to 300 observations across 6 predictors. The analysis uses cross-validated regularization to balance model complexity with predictive accuracy, selecting only the most influential channels while excluding noise.

Key Findings

  • R-Squared & Deviance Explained: 0.834 - The model explains 83.4% of sales variance, indicating strong predictive performance with meaningful channel effects captured
  • Variables Selected: 4 of 6 predictors retained - LASSO excluded 2 channels (predictor_4 and predictor_5) as non-influential, reducing model complexity
  • Prediction Accuracy: RMSE of 5.03 and MAE of 4.11 - Average prediction error is approximately ±4-5 units on the sales scale, with residuals symmetrically distributed
  • Lambda Selection: 0.661 (1se method) chosen over 0.085 (min) - Prioritizes stability and generalization over minimal training error, reducing overfitting risk

Interpretation

The model successfully identifies 4 advertising channels as meaningful sales drivers while eliminating 2 as redundant. The strong R² indicates these selected channels capture the essential sales dynamics. The conservative lambda choice (1se vs. min) suggests

Data Preparation

Data Pipeline

Preprocessing Summary

Data preprocessing and column mapping

Data Quality

Initial Rows300
Final Rows300
Rows Removed0
Retention Rate100

Data Quality

MetricValue
Initial Rows300
Final Rows300
Rows Removed0
Retention Rate100%
Processed 300 observations, retained 300 (100.0%) after cleaning

Interpretation

Purpose

This section documents the data preprocessing pipeline for the LASSO regression analysis. It shows that all 300 observations were retained without any rows removed during cleaning, indicating either pristine input data or minimal data quality issues. Understanding preprocessing integrity is critical for validating whether model performance (R² = 0.834) reflects true predictive power or data artifacts.

Key Findings

  • Initial Rows: 300 observations with no exclusions during preprocessing
  • Retention Rate: 100% — all records passed quality checks and remained in the final dataset
  • Rows Removed: 0 — no missing values, duplicates, or outliers were filtered
  • Train/Test Split: Not explicitly documented, though 10-fold cross-validation was used for lambda selection

Interpretation

Perfect data retention suggests the dataset arrived clean and complete, with no missing values or anomalies requiring removal. This is favorable for model stability but raises a subtle concern: the absence of any data cleaning may indicate either exceptional data quality or insufficient validation rigor. The LASSO model's strong performance (RMSE = 5.033, MAE = 4.113) is therefore more likely attributable to genuine predictive relationships rather than data artifacts.

Context

The lack of explicit train/test split documentation is notable; the analysis relied on 10-fold cross-validation for regularization parameter selection

Executive Summary

Summary

Key Findings & Recommendations

Key Metrics

r_squared
0.8341
rmse
5.0326
n_selected_vars
4
n_predictors
6

Key Findings

findingvalue
Model QualityGood fit (R² > 0.7)
Variables Selected4 of 6 predictors
Variables Excluded2 predictors set to 0
R-Squared83.4%
RMSE5.033
Optimal Lambda0.6614

Summary

Bottom Line: LASSO regression identified 4 of 6 predictor variables as relevant, achieving R-squared = 83.4% and RMSE = 5.033.

Variable Selection:
• 4 predictors have non-zero coefficients — these are the important predictors
• 2 predictors were shrunk to zero — excluded from the model by LASSO
• Lambda selection method: lambda.1se (= 0.6614)

Model Performance:
• R-squared: 83.4% of variance in the outcome explained
• RMSE: 5.033 average prediction error
• Deviance explained: 83.4%

Recommendation: Focus resources on the 4 selected predictors. Consider elastic net (alpha < 1) if predictors are highly correlated.

Interpretation

Purpose

This LASSO regression analysis evaluated 6 predictor variables to identify which ones meaningfully contribute to predicting the outcome while minimizing overfitting. The model successfully reduced the feature set through automatic variable selection, a key objective of LASSO regularization for improving model parsimony and interpretability.

Key Findings

  • R-Squared: 0.834 – The model explains 83.4% of variance in the outcome, indicating strong predictive performance with a parsimonious feature set
  • RMSE: 5.033 – Average prediction error of approximately 5 units, paired with MAE of 4.113, suggests consistent and reliable predictions across the dataset
  • Variables Selected: 4 of 6 – LASSO shrunk 2 predictors to zero coefficients, automatically excluding them from the final model
  • Lambda Selection: 0.661 (1se method) – Conservative regularization parameter chosen to balance bias-variance tradeoff, prioritizing stability over minimal training error

Interpretation

The model achieved strong explanatory power while reducing complexity from 6 to 4 predictors. The 1se lambda selection method prioritizes generalization over training fit, suggesting the selected variables are robust and unlikely to be artifacts of overfitting. All 300 observations were retained with no data loss, and residuals

Figure 4

Regularization Path

Coefficient Trajectories vs Lambda

Coefficient trajectories across the regularization path as lambda varies

Interpretation

Purpose

The regularization path visualizes how predictor coefficients shrink toward zero as the LASSO penalty (lambda) increases. This reveals the order and timing of variable selection—which predictors are most important (enter earliest at high lambda) versus least important (enter only at low lambda). Understanding this path is essential for identifying the core drivers of the model and validating the stability of selected features.

Key Findings

  • Lambda Range: 0.03 to 9.82 across 62 cross-validation folds, with lambda.1se = 0.661 chosen for regularization balance
  • Variables Selected: 4 of 6 predictors retained at the chosen lambda, with 2 excluded entirely
  • Coefficient Trajectory: Mean coefficient magnitude is 0.43 (sd=1.28), ranging from -1.82 to 3.07, indicating moderate effect sizes
  • Model Sparsity: Average of 3.76 non-zero coefficients per lambda value, confirming progressive variable elimination as penalty increases

Interpretation

The regularization path demonstrates that predictor_1 enters the model first (at highest lambda ~9.82), marking it as the strongest signal. As lambda decreases, additional predictors sequentially activate, with all 6 variables eventually entering at minimal penalty (lambda=0

Figure 5

Cross-Validation Error

Optimal Lambda Selection

Cross-validation error across lambda values with optimal lambda selection

Interpretation

Purpose

This section identifies the optimal regularization strength (lambda) for the LASSO model through 10-fold cross-validation. The lambda parameter controls the trade-off between model complexity and predictive accuracy—a critical decision that directly impacts which predictors are retained and how well the model generalizes to unseen data.

Key Findings

  • Lambda Min (0.085): Achieves the lowest cross-validation MSE (~24.76) but produces a more complex model with all predictors potentially active, risking overfitting.
  • Lambda 1SE (0.661): Selected for final model; provides comparable performance within one standard error of minimum while eliminating 2 predictors, yielding a simpler, more interpretable solution.
  • Cross-Validation Stability: MSE ranges from 24.76 to 153.38 across lambda values, with tight confidence bands (upper/lower bounds) indicating stable fold-to-fold performance at optimal lambda.
  • Final RMSE (5.03): Reflects the prediction error achieved using the selected lambda.1se regularization strength.

Interpretation

The analysis demonstrates a classic regularization trade-off: lambda.min minimizes error but at the cost of model complexity, while lambda.1se sacrifices minimal predictive power (~0.01 MSE difference) to achieve substantial simplification. This conservative choice aligns with the principle of parsimony—retaining only 4 of 6 predictors while maintaining strong cross-validated performance (R² = 0.834).

Context

The 10-fold cross-validation design ensures robust lambda selection across multiple data splits. The narrow confidence intervals at low lambda values suggest the model's performance is stable and reliable at the chosen regularization level.

Figure 6

Selected Coefficients

Non-Zero Variables at Optimal Lambda

Non-zero coefficients at the selected lambda — the variables chosen by LASSO

Interpretation

Purpose

This section identifies which variables drive the outcome and quantifies their individual impact. LASSO regularization automatically selected 4 of 6 predictors by shrinking irrelevant coefficients to zero, creating a parsimonious model that balances predictive accuracy with simplicity. Understanding coefficient magnitudes and directions reveals the relative importance and directional effect of each retained variable.

Key Findings

  • Variables Selected: 4 of 6 predictors retained (2 excluded by LASSO regularization)
  • Strongest Predictor: predictor_1 with coefficient +2.94 (largest positive effect on outcome)
  • Negative Effect: predictor_3 with coefficient −1.59 (only inverse relationship)
  • Weakest Selected: predictor_6 with coefficient +0.16 (minimal but non-zero contribution)
  • Directional Balance: 75% positive coefficients, 25% negative—outcome primarily increases with selected predictors

Interpretation

The model identifies predictor_1 as the dominant driver, followed by predictor_2 and predictor_3. The negative coefficient on predictor_3 indicates an inverse relationship: higher values decrease the predicted outcome. Predictors 4 and 5 were eliminated entirely, suggesting they add no independent predictive value beyond noise after accounting for the

Figure 7

Model Fit

Actual vs Predicted Values

Actual vs predicted scatter plot showing model fit quality

Interpretation

Purpose

This section evaluates how accurately the LASSO regression model captures the relationship between predictors and the outcome variable. Model fit quality is essential for assessing whether the selected variables (4 of 6) provide reliable predictions and whether the regularization approach successfully balanced complexity with accuracy.

Key Findings

  • R-Squared (0.834): The model explains 83.4% of variance in the outcome, indicating strong explanatory power across the 300 observations.
  • RMSE (5.033): Average prediction error is approximately 5 units, representing ~11.7% of the mean outcome value (43.01), suggesting reasonable practical accuracy.
  • MAE (4.113): Median absolute error of 4.1 units confirms consistent prediction performance without extreme outliers dominating the error distribution.
  • Residual Symmetry: Mean residual near zero (0.00) with minimal skew (-0.07) indicates unbiased predictions without systematic over- or under-estimation.

Interpretation

The model demonstrates strong fit quality, with the selected four predictors capturing the underlying data structure effectively. The tight alignment between R-squared and deviance explained (both 0.834) confirms the LASSO regularization successfully eliminated noise-bearing variables (predictor_4 and predictor_5) without sacrificing predictive power

Table 8

Performance Metrics

Model Fit Summary

Complete model performance metrics and parameter summary

metricvalue
RMSE5.033
MAE4.113
R-Squared0.8341
Deviance Explained0.8341
Lambda (1se)0.6614
Lambda (min)0.0854
Variables Selected4
Total Predictors6

Interpretation

Purpose

This section provides a comprehensive snapshot of the LASSO regression model's predictive performance and feature selection efficiency. It answers whether the model achieves adequate accuracy while maintaining interpretability through automatic variable selection, which is central to the analysis objective of balancing prediction quality with model simplicity.

Key Findings

  • R-Squared & Deviance Explained: 0.834 - The model explains 83.4% of variance in the target variable, indicating strong predictive power across the 300 observations with no data loss.
  • RMSE: 5.033 - Average prediction error magnitude; paired with MAE of 4.113, suggests relatively symmetric error distribution with minimal outlier influence.
  • Feature Selection Efficiency: 4 of 6 predictors selected - LASSO successfully eliminated 2 non-informative variables (predictor_4 and predictor_5), reducing model complexity by 33% while maintaining performance.
  • Regularization Parameter: Lambda.1se = 0.661 chosen over lambda.min (0.085), prioritizing stability and generalization over training fit.

Interpretation

The model demonstrates robust performance with strong explanatory power and effective dimensionality reduction. The gap between lambda.min and lambda.1se selection reflects a conservative regularization strategy that trades minimal training improvement for substantially improved generalization potential.

Want to run this analysis on your own data? Upload CSV — Free Analysis See Pricing