Prediction of Auto Insurance Risk Based on t-SNE Dimensionality Reduction

Correct risk estimation of policyholders is great significance to auto insurance companies. While the current tools used in this field have been proven practice be quite efficient and beneficial, we argue that there still a lot room for development improvement process. To end, develop framework based on combination neural network together with dimensionality reduction technique t-SNE (t-distributed stochastic neighbour embedding). This enables us visually represent complex structure as two-dimensional surface, while preserving properties local region features space. The obtained results, which are real data, reveal clear contrast between high low policy holders, indeed improve upon actual performed by insurer. Due visual accessibility portfolio approach, could advantageous insurer, both main prediction tool an additional validation stage other approaches.