Use case stb
1 like180 views
1) The document discusses using machine learning and AI in banking and finance, specifically for a case study on a Tunisian bank (STB Bank). 2) It outlines several potential use cases for machine learning in areas like credit scoring, fraud detection, customer sentiment analysis, risk management, and operational efficiency. 3) The methodology discussed follows a CRISP process involving business understanding, data understanding, data preparation, modeling, evaluation, and deployment with feedback.
Use case stb
- 1. Use case : Machine Learning and AI in banking and fnance Dr Ahmed Rebai Assistant Professor Of Data Science Esprit School of Engineering 29 December 2018 Dr Lotf Ncib Assistant Professor Of applied mathematics Esprit School of Engineering
- 2. 1 2 3 4 5 6 Table of Contents Introduction Retrospectives STB Bank Use Case Presentation Data science methodology Project’s steps Conclusion 2 Selecting DS methodology – Available data Business understanding- Data's Phases– Modeling-Evaluation-Deployment-Feedback Ahmed Rebai-Lotf cib
- 3. Introduction 3 Business Banking and Finance solutions Credit ranking system... Data Varity – Volume – Digitalization Business Intelligence Dashboarding – Intelligent visualization Data science Exploitation-Meaning-Prediction Ahmed Rebai-Lotf cib
- 4. 4 public & semipublic sector (52.5%) foreign actors (11.3%) private sector (36.2%) Financing of Industrial companies 18% of local Market part Investment banking Founded in 1958 In Tunis bank's capital participation CEO Monsieur Samir Saied Various instruments Ahmed Rebai-Lotf cib
- 5. CRISP Methodology 5 Business Understanding 1st Phase Data Understanding 2nd Phase Data preparation 3rd Phase Deployment Final phase Evaluation 5th phase Modeling 4th phase CRISP Methodolog y Ahmed Rebai-Lotf cib
- 6. 1 2 3 4 The Master Plan 6 The new Data Science Methodology – (IBM vision 2018) How can you use data to answer the question? Analytic Approach What data do you need to answer? Data requirements Where is the data coming from and how will you get it? Data Collection Can you get constructive feedback into answering the question? Feedback Ahmed Rebai-Lotf cib
- 8. Tools that we will use 8 ETL + Reporting Pentaho Data Integration Dbeaver – PHPMyAdmin => MySQL database Studio3T => MongoDB database Power BI Linux Data Science Python (numpy, pandas, matplotlib, sklearn, tensorfow, keras, pytorch, textblob, senpy, nltk,...) Google Cloud Microsoft Azur Amazon WebServices Ahmed Rebai-Lotf cib
- 9. Business Understanding 9 Fraud Detection Customers Sentiment analysisAI & ML are used to identify sentiments in textual data: in social media comments, news articles . Risk Management Operational efficiency: i ML and Graph theory can detect pattern towards fraudulent operartions (see Panama papers case HSBC Bank) ML can predict risk arising out of banking exposures. Risk could be either credit risk or fraud risk from transactions or specifc customers. A simple use-case is to convert hand-written forms into machine readable data. This helps in reducing costs signifi-cantly as most banking processes require lot of paperwork. Ahmed Rebai-Lotf cib
- 10. Analytic Approach - P1 10 Semi-structured data contains : Clients’ information “Agences bancaires” ’ information DABs’ information Transactions’ information Find relation between clients and DAB in Transactions data. Week relationship between “Agences bancaires” and Transactions. How can you use data to answer the question? Develop a datawarehouse with this available data and try to centralize the information in order to have a clear idea in Modeling phase Ahmed Rebai-Lotf cib
- 12. 1 2 3 4 5 Modeling – P2 12 Type of model : Supervised method Algorithm: ARMA, ARIMA , SARIMA , SARIMAX, Implementation : Python Robustness & Evaluation = Stochasticity evaluation , Rsquared and Accuracy AIC Detection of Trend , Seasonality + residuals evolutions Users 'number forecasting Ahmed Rebai-Lotf cib
- 13. 1 2 3 4 5 Modeling – P3 13 Type of model : Unsupervised method Algorithm: CAH , KMEANS, Dbscan Implementation : Python: sklean, Tensorfow Robustness & Evaluation = silhouette score Providing the clusters of users and then using them for group charact-erization Users’ profling Ahmed Rebai-Lotf cib
- 14. 1 2 3 4 5 Modeling – P4 14 Type of model : Supervised method Algorithm: LDA , Logistic regression Implementation : Python Optimization & selecting model = GREEDY Wilks Setting a score for each Reward / Loyalty based on the number of transactions Reward/Loyalty Scoring Ahmed Rebai-Lotf cib
- 15. 1 2 3 4 5 Modeling – P5 15 Type of model : Supervised method Algorithm : NLP , Stemming , lemmatization Implementation : Python Robustness & Evaluation = MDT , IDF Detect word weights that attract users Knowledge text discovery Ahmed Rebai-Lotf cib
- 16. 1 2 3 4 5 Modeling – P6 16 Type of model : Supervised method Algorithm: collaborative fltering , Turicreate , CF Implementation : Python Robustness & Evaluations = RMSE , NDCG , Mean Reciprocal Rank Recommend a fnancial product (specifc category) in a specifc period , in a specifc region Recommend a user for a loyalty ofer. Recommender system Ahmed Rebai-Lotf cib
- 17. 1 2 3 4 5 Modeling – P7 17 Type of model : Supervised method Algorithm : Decision Tree , Random Forest Implementation : Python Robustness & Evaluation = Roc Curve , Accuracy Detect the conditions to take a ofer or not Need external tracking data of users in the web application: Page views , clicks… Boosting with Random Forest Ahmed Rebai-Lotf cib
- 18. 1 2 3 4 5 Modeling – P8 18 Type of model : Unsupervised method, Graph theory, discrete mathematics. Algorithm : Clustering, Community detection, Outliers detection Implementation : Python Robustness & Evaluation = Roc Curve , Accuracy Detect suspicious operations Fraud Detection Ahmed Rebai-Lotf cib
- 19. Deployment 19 With Juputer-lab or External Web site Ahmed Rebai-Lotf cib
- 20. Feedback 20 Can you get constructive feedback into answering the question? Ahmed Rebai-Lotf cib
- 21. “In God we trust, all others must bring data.” W. Edwards Derming Ahmed Rebai-Lotf cib
- 22. Thank you for your attention 29 December 2018 Ahmed Rebai-Lotf cib