An Intelligent Hybrid AI–TOPSIS Evaluation Model Using Machine Learning, Deep Learning, and Natural Language Processing on Databricks for Open Banking

Abstract

This paper proposes a hybrid evaluation framework combining Artificial Intelligence (AI) and the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) to assess electric motorcycle performance within emerging open banking ecosystems. The model integrates traditional multicriteria decision-making (MCDM) with supervised machine learning and deep learning algorithms to produce robust, data-driven rankings of electric motorcycle alternatives. Data sources include telemetry and sensor logs, consumer finance and usage patterns from open banking APIs, and lab-based performance tests. Preprocessing and feature engineering are performed within Databricks to exploit cloud scalability and distributed processing, enabling efficient handling of large, heterogeneous datasets. Machine learning components—random forests and gradient boosting—provide feature importance and predictive baselines for key performance indicators (KPIs) such as range, energy efficiency, acceleration, reliability, and total cost of ownership. Convolutional and recurrent neural networks are applied for time-series and image-derived features (e.g., thermal maps, battery degradation signatures). The hybrid pipeline uses model outputs to weight and normalize criteria for TOPSIS ranking; specifically, machine-derived importance scores inform objective weighting while domain experts supply subjective weights, producing a hybrid weight vector. The TOPSIS stage computes closeness coefficients to ideal solutions, yielding a final ranking and sensitivity analysis. Experimental evaluation on a synthetic-but-realistic dataset and a case study with three commercially available electric motorcycles demonstrates improved ranking stability and predictive accuracy compared to baseline MCDM and ML-only approaches. The framework further illustrates how Databricks' scalable environment and integration with open banking APIs facilitate near-real-time assessment for stakeholders—manufacturers, consumers, and financing partners—enabling data-driven purchasing and financing decisions. Limitations, policy implications for open banking data privacy, and directions for future research are discussed.