140 likes | 332 Views
Consultancy Project Preliminary Findings . Using Data Mining to Identify TNB Customers Likely to Default Payment. Group Members for this project are from the COIT, UNITEN. Alan Cheah Kah Hoe (Leader) Assoc. Prof. Dr. Mohd Sharifuddin Ahmad Mohana Shanmugam Zaihisma Che Cob
E N D
Consultancy ProjectPreliminary Findings Using Data Mining to Identify TNB Customers Likely to Default Payment
Group Members for this project are from the COIT, UNITEN • Alan Cheah Kah Hoe (Leader) • Assoc. Prof. Dr. Mohd Sharifuddin Ahmad • Mohana Shanmugam • Zaihisma Che Cob • Mohammad Shukeri Yusuff • Ammuthavali Ramasamy Kicked-off date :15th October 2012 Duration : 9 months Expected Completion date : 15th July 2013 Cost of Study: RM99,875
CRISP-Data Mining Methodology: a powerful tool to detect trends and patterns using data Cross Industry Standard Process for Data Mining is a methodology process that is used to mine huge data. E.g. e-CIBS Hidden trends and patterns will be identified using CRISP-DM. Main research question: “To identify customers who are likely to default payment to TNB Distribution
Data Understanding- raw data from eCIBS TNB Bangi and close collaboration with Dist Finance….we categorised the customers’ credit worthiness • Data from Station 180 Bangi extracted from e-CIBS : • Customers Data (210,000 records) • Customers Payment History (Jan -Oct 2012) • Categorized into Credit Worthiness 1 and Credit Worthiness 345 • Credit Worthiness is used to evaluate the customer’s credit rating based on criteria and factors defined by TNB. • Each customer would be assigned one of the following credit ratings:- 0 - Excellent 3 – Below average 1 – Above average 4 – Poor 2 – Average 5 – Very Poor
Data Collection – in addition to raw data from eCIBS, we collected data using survey for demographic and behavioural information…. • Conducted questionnaire survey for customers demographic and payment behaviour data. • Survey conducted : • Kedai Tenaga Bangi (759 records) • Kedai Tenaga Kajang (825 records) • Online Survey (in progress)
Modelling-preliminary results indicate links of credit worthiness with some factors as shown …..however more detail analysis still needed • The model has successfully predicted the type of customers based on the questionnaire survey and customers payment history data. • Questions that contributed significantly are:- • Account Holder or Tenant • Age Group • Employment • Type of Premise • Reasons for non-payment : Out of country
Modelling – Prediction of Good, V.Good,Bad and V.Bad Customers Accuracy is 99.5% ..only 3 records were predicted wrongly
Findings – Category of Good/Bad Customers ….domestic customers prevail as not good paymasters
Findings – Kedai Tenaga & Pos Malaysia still No.1 & No.2 in payment channels…online payment channel usage is still low in TNB Bangi
Findings – Awareness of Online Payment e.g. TNB e-services…can we infer that if TNB increase promotion of online services, then there will be less bad paymasters
Moving Forward • To deploy the model on new sets of data in evaluating the accuracy of the model (data from survey, site as well as online) • New payment history data to be obtained from e-CIBS.
Conclusion • The project is progressing well and on schedule. • The model will determine the factors that affect the customers’ payment to TNB • Good cooperation from SMOD & e-CIBS team • Responses through online survey are slow