Private Banks in Bangladesh has been facing serious challenges since the very beginning in terms of iden-tifying appropriate loan applicants. Various factors such as growing inflation, property price rise, nepotism and biasness, personal networking have made the application process highly inaccurate for proper identi-fication (Ziaul, 2003) of loan applicants. The NPL (non-performing loans) within the growing banking sector of Bangladesh since the beginning of financial flexibility in Bangladesh has played as a positive inhibitor for financial independence (Rezina, 2020) but failed to effectively high-light correct loan provision policy. Although lots of surrounding political and environmental factor plays a crucial role for NPL growth but the majorities of growths are still inscribed within the human sentiments (Ali, 2012). In this research paper we collectively focus on factors that are associated with human faults rather than honest mistakes in order to make the research process effec-tive. Our primary analysis that was raised through Ziauls question refers to determining variable cri-terions of the loan applicants and effective data analysis for suggestive solution formation. Vedala et al. have found that Naïve Bayes classifier is an effective method for identifying good and bad borrowers for better financial decisions (Vedala et al., 2012). In order to understand the possible criterions to judge the loan applicant factors related to their current situation needs to be stipulated (Krichene, 2017). Although Rezina already mentioned that in Bangladesh despite of growth there are various identifiable factors those needs to be judged in order to make the loan provision process acceptable, but Bangladesh Bank has yet to make proper adjustments accordingly. 

Banks overall financial development through this process needs to be an important indirect factor for loan approval. Sami et al. (2021) have shown that how financial ratio becomes an important inhibitor for asset selection in case of financial investment (Sami et al., 2021). Its also shown by HM Sami that in case of effective asset selection both financial ratios and classification method seems important (Sami, 2021). This research process will associate banks financial performances as an important background justifying classification process as an inhibitory action for selection of borrowers. In order effectively understand the criterions for selection various similar situation sorts needs to be analyzed and justified as important criterions for loan approval (Majeske, 2013). 

In order to understand the financial implications of the financial viewpoint, Bangladesh Bank has governed several parameters to ideally identify a potential bor-rower (Weber et al., 2015). Weber et al. have further evaluated the strengths of the given criteria as im-portant standpoints for financial sustenance through loan provision method for banks. Hence in this research process we will use successful classification process to clearly allow only definitive criterions to check the financial perseverance of the borrowers for selected business plans. In general, our overall process is highly data supported and variable oriented hence we didnt propose any deep differentiation of loan types, fund requirements and business types associated with fund requirements as an important benchmark for financial analysis. 

Literature Review

Singh et al. (2021) have figured out that through modernized methods of loan approvals based on machine learning parameters such as XGBoost and Random Forest process of decision tree have clearly considered credit score and other parameters as im-portant decision-making parameter for loan approval. Similarly, its also seen that through financial trans-actions credit scores could be determined which are important basis for financial decision makings (Loho-kare et al., 2017). Its seen in the research process of Ziauls suggestion, although surrounding environ-mental factors plays a crucial role in loan approval decision both from borrower and lender perspective but this research would concentrate specifically on the financial aspect terms with association of observable factors as important borrowing determinants (Kemal-bay et al., 2014). Kemalbay et al. also critically suggested that in case of minorities with low income and shorter employment history has higher loan default rate along with higher rates of disapproval (Kemalbay et al., 2014). Calem has made an effective suggestion that shows the true relative analysis of income as benchmark for loan approval and promissory loan return (Calem, 1996). Squires & Kim, (1995) suggested in their primary overview, with effectivity in terms of gender, minority and income level loan approvals remain an effective classification parameter. Its also seen that with increasing development around the world and also with more ease and effectivity in reducing discriminating parameters more subjective issues became better determiner for loan approvals such as income level and employment history rather than discriminant factors (Hassan et al., 2021; Harrison & Glover, 2008). 

A major segregation or classification comes around the factor of financial condition of the applicant with reference to their asset condition and liquidity of coll-ateral in reference to loan. There are various financial stages which generally classify a person as a better borrower than another (Carter et al., 2007). With NPL rise in Bangladesh although Rezina clearly suggested a stricter approach towards approval but Naïve Bayesian methods will provide a definitive benchmark for appli-cants eligibility through past datas (Rezina, 2020). In reference to information available for the public, financial ratios as performance benchmark plays an important role for judgment of the banks present per-formance standards in terms of EPS & PE Ratio (Sami HM, 2021).  Its also seen that in Bangladesh especially for private banks the performance measurement factors are highly reliant on financial ratios such as EPS and PE Ratio (Siddik et al., 2017). Mayer & Gary has identified that in case of making loan approvals, most defaulters tend to set a smaller down payment boundary which seems more effective towards default rather a larger payment always makes the default risk more mitigated (Mayer & Gary, 1996). Loutskina & Strahan has made further contributions towards the field of available liquidity of borrowers assets as an important background loan approval (Loutskina & Strahan, 2009). Although most operating industries effectively applies for loans using project profile and previous work history but the effective results need to be portrayed in financial ratios for performance develop-ment or evaluation (Sami, 2021). Referring to the length of the loan term associated with loan approval Heider & Inderst made the most appropriate sug-gestions. Its found that on the basis of length of the loan terms a higher interest rate seems more favorable for the loan because of increasing inflation (Heider & Inderst, 2012). Kathleen & Li demonstrated that per-sonal debt and income ratio plays an important criterion for selection of borrowing individuals (Kathleen & Li, 2010). In relation to finding equality it seen that people with higher incomes generally tend to be better bor-rowers for banks in reference to same debt levels in terms of money (Kathleen & Li, 2010). But in general, lower debt to income ratio remains a favorable back-ground for banks in order to identify a better borrower. Bian et al. have found that with smaller deposits on loan or less liquid assets as collaterals the amount of loan to the original property value for loan borrowing becomes highly close to 1 which remains a signal for bad borrower (Bian et al., 2018). Blackwell & Winters have found out that if a borrower with more assets rather than their liabilities then its more favorable for banks to make lendings for them (Blackwell & Win-ters, 1997). Its also seen that the best possible loan default prohibitions happen only against greater than 1 asset liability ratio.

Theoretical Framework 

Debt to Income Ratio: The debt-to-income ratio (DTI) is a lending ratio that represents a personal finance measure, comparing an individuals debt repayments to his or her gross income on a monthly basis. Gross income is simply a monthly paycheck before one pays off the costs, such as taxes, interest expense, etc.

Debt-to-Income Ratio = Total Monthly Debt Payments/Gross Monthly Income

In the research perspective, for the associated benefits Bangladesh bank have arranged a guideline to govern each banking activities as subordinating regulation by which their regular debt payments are supplemented by regular monthly payments against any loans. Under this perspective the newly formatted Debt to Income Ratio if greater than 1 then the banks are at risk and if lower then the payments are smooth. Hence the reformulated formula is: 

Debt to Income Ratio = Total Monthly Debt Balance for Loans/Total Monthly Loan repayments

Similarly, the effectivity of Loan to Value Ratio plays an important role for loan approvals. It seems that 

Loan to Value = Amount of Loan / Property Value

In reference to the approval of loan if the ratio is very close to 1 then the possibility of approval should be very limited whereas in case when the loan is very close to 0 then approval should be easier. The bor-rowers notion to current ratio will have to modified for loan approval view. In accordance to company-based scenario if an organization has more than 1 current ratio then the organization has more assets than its liabilities. Similarly, if a borrower has more assets rather than liabilities through asset collaterals or by financially liquidating sources. So, working capital ratio or current ratio for loan borrower is 

Current Ratio = Current Assets Owned/Current Liabilities as loans

In this research we would arrange a framework to identify applicants with above mentioned ratios and use various statuses including credit status, income status, networking abilities, employment length, collateral liquidity, down payment size, number of loan terms and loan length plays vital role in identifying a qualified borrower. The three financial variables will indicate the financial conditions of the borrower whereas the other variable features from the perspective of previous performance of the borrowers would generate the bets possible notion of borrowers behavior in reference to lending. 

Research Framework

In order to make this process effective, in this research we have initiated the primary process of selecting applicants from 5 different banks. Each of these banks have a NPL of more than 30% and weaker payment pattern of 56%. Most of these banks have extremely large capital but the loan approval process was highly non-conforming. Using K Means clustering we have been able to distinguish bad chunk of loan applicants pretty easily using the parameter setup through algo-rithmic boundaries. With the usability of high net-working parameters, low debt to payment ratio, low loan to value ratio and more than 1 current ratio of the borrowers, the primary borrower background should be selected. After making the selection of these borrowers, using these borrowers previous loan payment pattern as training data, a section of testing was arranged as a pilot process for major evaluation by naïve bayes classifier to judge the overall criteria that could be accepted from a group of selected applicants for loan approval. 

Fig 1: Outline of the Research plan.

Using K Means Clustering we would select applicants by following algorithmic features: 

Xi = Networking &Social status 

Yi = Loan to Payment Condition 

Zi = Loan to Value 

Ki = Current Ratio of Borrower 

For loan applicants to be acceptable in accordance to the issue of bank approval for lending to a borrower of effective acceptability various deciding factors are followed –

Xn = High Networking & Social Status 

Yn = High Loan to Payment Ratio 

Zn = Low Loan to Value Ratio 

Kn = More than 1 Current Ratio of the Borrower

Using these factors 

Acceptable Terms are generated as: 

Acceptable Borrower = Min (Abs (Xn – Xi) + Abs (Yn – Yi) + Abs (Zn – Zi) + Abs (Kn – Ki))

Only the Lowest possible values would then be accepted. After the selection of primarily selected borrowers with good loan paying status, we would then extract data of similar features from the previously applied loan applicants. Its been seen and observed that various loan paying individuals are capable of various positive features but has exhibited under pay-ment or no payment despite of positive features. Through Naïve Bayes Classifier these data are the further evaluated for finding out effective loan appli-cants. While proceeding with Naive Bayesian classi-fication process we need to administer various back-ground to develop its importance. 

Where each of the criterions are separately calculated on the basis of given probability.

Here all the criterions are being exhibited together.

If all the probabilities of the events are justified are denoting a joint probability due to being mutually exclusive, hence the Probability of lender satisfaction being earned given the events are contained is directly proportional to the probability of satisfaction multiplied with probability of satisfaction earned when each event are satisfied.

In the final step of argmax, its clearly denoting if any specific criteria remain positively influencing towards a pathway of decision, then it would probably denote the definitive answer. Hence by this process the selected pool of applicants within common boundary criteria could be selected with positive and effective traits. All the variables of xi refer to the exhibiting criteria from feature set that would be defined under various strength and week parameter. Generally, from a lenders pers-pective the strength parameters make the suitable decision for loan approvals. Among the selected more than 3000 applicants of different status and nature from of pool of 10000 applicants for the term of 2021 January- March session, we were able to easily classify best applicants using K Means Clustering. Among the 3000-applicant pool in various private banks there were quite a lot of them with good features but those features didnt make good examples previously. After extracting previous data of 12343 applicants of similar features it was found that nearly 4678 of them with similar fea-tures were running underpayment or didnt complete loan payments for the banks. Hence it was of para-mount importance to understand definitive features by which a loan applicant will be able to make payments properly and who will not be able to do it. 

Fig 2: High Network & Social Status.

Fig 3: Low Debt to Payment Ratio.

Fig 4: Income Status.

Fig 5: Employment History.

Fig 6: Down Payment Size.

In order to effectively segregate the selected applicant, pool its been found that with the above average range of expected growth from the minimum till 1 there has been a credibility of return of more than 75% with pro-per repayments and no late or underpayments. Hence 

[P(High Network & Social Status > 0.6)] = P(Xa) = 1 – P(~Xa)

[P(Low Debt to Payment Ratio > 0.4)] = P(Ya) =

1 – P(~Ya)

[P(Income Status > 0.4)] = P(Za) = 1 – P(~Za)

[P(Employment History > 0.4)] = P(Wa) = 1 – P(~Wa)

[P(Down Payment Size > 0.4)] = P(ka) = 1 – P(~Ka)

Algorithm 

So,  if 

Results Analysis: 

Our results are demonstrated under two different parameters. Primarily we were capable of successfully segregating the borrower group of better promise and capability. Then among those 3000 capable loan appli-cants we were able to select the best 2133 applicants based on features which all exhibited positive notions of repayment. In this process for January till March, the Bank has found that around 98.3% of the selected applicants were making regular payments with only 4 people making no payments in 3 months interval. 

Fig 7: Results Analysis.

Scopes of Improvement & Limitations

This research is conducted strictly under the notion of statistical dependency to construct a definitive back-ground for eligibility of loan candidates. But the overall process of Loan approval needs to highlighted under the context of fundamental business development process along with various scopes of business growth and proposition, hence this research process needs to evaluated under fundamental perspective of under-standing loan approval processes. Similarly, various benchmark features suggested by various other resear-ches are considered as pioneering variables in this research process for determining loan approval decision but in reality, various other variables could had been evaluated and judged for better performance appraisals. 

Finally, the process is high degree reduction policy which does not invite more investment opportunity and is very suitable for economies with tightening policies as whole but does not exactly exhibit the effective nature of finding out true qualifiers for Banks and furthermore economic benefit. 

Finally, the process is high degree reduction policy which does not invite more investment opportunity and is very suitable for economies with tightening policies as whole but does not exactly exhibit the effective nature of finding out true qualifiers for Banks and furthermore economic benefit. 

For Bangladesh with growth of 5.5% inflation and around 7% GDP growth rate this process of stringent loan approval could be highly useful. Although its highly statistical but such process would eliminate applicants of low-quality return capability. Albeit its observed that in our research process we lack the evi-dence of support newer or potentially capable firms in many ways but this has reduced the loss of loan default to less than 1% which could be an extremely good method of support. 

We do acknowledge the continuous support and help from various privatized banks who have helped us relentlessly through data provision and continuous support with managerial application changes for the application of such suggestive reforms. Due to Market conflict the names of such private organizations are remained anonymous in this research process. 

We declare no single potential conflict of interest.