Machine learning uses algorithms to map input data to predictions. In tabular-data work, the term commonly means classical approaches rather than neural networks, spanning linear regression through gradient boosting. Its practical strength comes from turning business data into useful features, then fitting a model to the resulting rows and columns. section 1.3 section 1.6
Feature engineering is central: select relevant information, discard noise, represent values in forms a model can use, address nonlinear relationships, and incorporate domain knowledge. These transformations help a model learn patterns that generalize beyond the training data. section 1.3
For tabular business problems, assess machine learning against deep learning on three practical criteria:
- Simplicity: prefer a solution with less complex application code and a manageable framework API. section 3.0 section 3.5
- Transparency: favor models whose behavior and feature effects can be explained to business stakeholders and, where needed, regulators. section 3.2.3 section 3.5
- Efficacy: compare predictive results along with training and implementation time; out-of-the-box results may be too close to declare a winner. section 3.0 section 3.5
Airbnb price classification
A concrete tabular classification task predicts whether a New York City Airbnb listing’s price will be above or below the market average. An XGBoost solution represents machine learning and can be compared with a Keras functional-API deep-learning solution for code complexity, transparency, and efficacy. section 3.1 section 3.0
Gradient boosting tools such as XGBoost or LightGBM are conventional choices for tabular data, but deep learning remains worth including in an analysis toolkit. Choose from evidence for the specific dataset and operational priorities rather than treating either approach as universally best. section 1.6 section 3.0
