RamKumar M.

Idea is to Analyze the unstructured data and identify the critical clauses / Phrases which point the complaint issues, where companies can pinpoint and rectify the issues or use Preventive Mechanism. Techniques Used: • Vectorization method used: TF – IDF (term frequency, inverse document frequency) 3 • Clustering Method – TF IDF Clustering. We cannot directly visualize TD IDF since the data is in sparse matrix and require dimension reduction, as TF IDF clustering is not a common clustering method. • Conversion of the TF-IDF score into format which has (X, Y) coordinates. Truncated SVD - which is linear dimension reduction by Singular Value Decomposition - Sparse matrix into A dense matrix. - Still we cannot visualize since it has higher dimension with random Probability of occurrence. In simple words, the data is Stochastic. - We need to use t-SNE method. (t – distributed Stochastic Network Embedding) - From this analyze, we can identify some of phrases which can have impact - Used K Mean Clustering as secondary method.

Large Multinational Financial Company wants to optimize its customer remediation process by incorporating Machine learning. The company needs to identify the impacted customers on a regular basis for a given issue and to predict the appropriate remediation amount to be paid. As part of the project, I built a Machine learning model using ensemble techniques on random forest and gradient descent methods to identify the impacted population and to predict remediation amount range. This model on subsequent training and tuning yielded 91 % accuracy in predicting the amount range and was fully incorporated in the client process.