b. Bala ChandraIndependent in spirit. Engineer by craft.
Let’s talk
All selected work

Case study 04 — Machine learning research

DemandForecasting.

A reproducible NYC taxi demand pipeline that moves from raw movement data to compared models and interpretable predictions.

Domain
Transportation ML
Data
NYC taxi demand
Models
XGBoost · LSTM · TCN
Method
Preprocessing · Features
Explain
SHAP interpretation
View repository
Fig. 04 / PipelineRepository
01Taxi data
02Preprocess
03Features
04Model compare
05SHAP
01Question

Can city movement become a useful forecast?

The project studies transportation demand as a structured prediction problem, using NYC taxi activity as the basis for a complete modelling workflow.

It deliberately covers more than model fitting: the upstream decisions that shape the signal, and the downstream work needed to compare and explain what a model learned.

02Method

Build the evidence chain.

01

Prepare

Clean and organise trip data so temporal demand patterns are represented consistently.

02

Engineer

Aggregate and construct features that turn raw activity into model-ready demand signals.

03

Compare

Evaluate XGBoost, LSTM and TCN instead of assuming one model family wins.

04

Interpret

Use SHAP to connect predictions back to understandable drivers.

03Workflow

From raw trips to interpretable demand.

  1. Ingest taxi records
  2. Preprocess and aggregate
  3. Engineer features
  4. Train and compare models
  5. Interpret with SHAP

Evaluation note: the model families and interpretation method are documented, but no performance figures were published for this write-up, so no accuracy or ranking claim is made.

04Takeaway

Forecasting is a system, not a score.

Data preparation, temporal modelling, comparative evaluation and interpretability belong together — that combination is what makes mobility work trustworthy.