Scalable Open Banking and Electric Mobility Analytics: Machine Learning and Deep Learning Integration with APIs, TOPSIS Optimization, and Cloud-Based Databricks AI

Abstract

Open Banking and Electric Mobility (e-mobility) represent two rapidly converging domains where data-driven analytics can deliver substantial societal and commercial value. This paper proposes an integrated framework that combines transactional Open Banking data and e-mobility telemetry through secure APIs, applies machine learning (ML) and deep learning (DL) techniques for predictive analytics, and uses TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution) as a multi-criteria decision support layer to prioritize operational actions and deployment of services. The cloud layer is implemented on a modern Data + AI lakehouse (Databricks) to provide scalable ingestion, feature engineering, model training, and model serving with governance and lineage. In the financial dimension, ML models detect transactional patterns, creditworthiness shifts, and real-time micro-product recommendations derived from account-level signals. In the mobility dimension, vehicle telemetry (charging, range, battery state, geolocation) is fused with grid and pricing signals to provide energy-aware routing, demand forecasting, and battery health estimation. Models include gradient-boosted trees for tabular financial tasks, CNN/LSTM hybrids for time series and telemetry, and graph embeddings for customer-vehicle-infrastructure relationships. TOPSIS is applied to rank candidate interventions (e.g., charging station deployment, targeted financial offers, demand response actions) across criteria: expected revenue uplift, customer impact, operational cost, regulatory risk, and carbon reduction. Implementation on Databricks demonstrates near real-time pipeline throughput, reproducible ML lifecycle (MLflow), and secure API integration with OAuth2 / OpenID Connect for consented data access. Results on a combined synthetic + anonymized dataset show improvements in forecasting accuracy (MV forecasting RMSE reductions of 12–18%) and a TOPSIS-driven prioritization that increased expected utility by ~22% versus single-metric baselines. The paper discusses system design, privacy and regulatory considerations, and recommendations for scaling to production. (European Central Bank)