How Credit Scores Estimate Risk
Photo: Towfiqu barbhuiya / Pexels
A FICO score, the most widely used credit scoring model in the United States, ranges from 300 to 850. That range was not picked for aesthetic reasons. The model behind it was built from historical repayment data on large numbers of borrowers, so the number works as a statistical estimate of default risk: the likelihood that a borrower will fail to repay a debt as agreed. This article explains how that number gets built, what goes into it, and where it falls short. It is educational information, not financial advice, and it does not tell you how to change your own score.
What a credit score actually is
Credit score: a three-digit number, typically ranging from 300 to 850 depending on the model, that summarizes information from a consumer's credit report into a single estimate of default risk. It is not a measure of income, net worth, or savings. A person earning a high salary can have a mediocre score, and a person with modest income can have an excellent one, because the score reflects borrowing and repayment behavior specifically, not overall financial standing.
The simplest way to think about it is as a forecast. A weather forecaster who says "70 percent chance of rain" is not claiming it will rain tomorrow. The claim is that across many days with similar conditions, it rains on about seven of every ten. A credit score works the same way. It compresses a borrower's history into a rank, and the rank corresponds to how often borrowers with similar histories went on to miss payments. For any single person the forecast can be wrong, even when it is accurate on average.
A credit report and a credit score are different things. The report is the underlying record, like a school transcript: a file maintained by each of the three major bureaus (Equifax, Experian, and TransUnion) listing a consumer's open and closed accounts, payment history on each, amounts owed, and inquiries from lenders who have checked the file. The score is the grade point average computed from that transcript. Different scoring models can read the same report and produce different numbers, which is a theme this article returns to below.
A short history of scoring
Fair Isaac Corporation, the company behind the FICO score, was founded by engineer William Fair and mathematician Earl Isaac, on the premise that data analyzed systematically could improve lending decisions. It built scoring systems for individual lenders, starting in the late 1950s, before the 1989 general-purpose score. Before general-purpose scores existed, lenders often relied on manual underwriting, in which a loan officer reviewed an application and used personal judgment to approve or deny it.
FICO introduced its general-purpose credit score in 1989, and the score's use spread widely in the mid-1990s, when Freddie Mac required credit scoring for new mortgage applications (Wikipedia, FICO Score). Different versions have followed, as described below.
What actually feeds into the score
FICO has publicly disclosed the approximate weighting of the five categories that make up its most widely used general-purpose scoring model. The breakdown below follows myFICO, the company's consumer education site (myFICO, What's in my FICO Scores).
| Factor | Approximate weight | What it measures |
|---|---|---|
| Payment history | 35% | Whether payments on credit accounts were made on time, and how late or how often they were missed |
| Amounts owed | 30% | Total debt across accounts, and credit utilization, the share of available revolving credit currently in use |
| Length of credit history | 15% | Age of the oldest and newest accounts, the average age of all accounts, and how long since certain accounts were used |
| New credit | 10% | Recently opened accounts and hard inquiries (credit checks made when you apply for credit) |
| Credit mix | 10% | The variety of credit types on file, such as credit cards, retail accounts, installment loans, and mortgages |
Payment history and amounts owed together account for roughly two-thirds of a typical FICO Score, according to myFICO's published methodology.
Payment history carries the most weight because a track record of on-time payments has historically been one of the strongest available predictors of future on-time payments. This is intuitive: past behavior, aggregated across millions of accounts, correlates with future behavior more reliably than almost any other single factor in a credit file. myFICO adds that a few late payments are not an automatic "score-killer," and that how recent, how severe, and how frequent the lateness was all matter.
Credit utilization, the percentage of available revolving credit currently being used, sits inside the amounts-owed category. Consumers who consistently use a high share of their available credit have historically defaulted at higher rates than those who use less, according to FICO's published research. The mechanism is not moralistic. It reflects a statistical pattern observed across large populations, not a judgment about any individual borrower's character.
What the score never sees
The list of excluded information is as informative as the list of inputs. myFICO states that FICO Scores do not consider race, color, religion, national origin, sex, marital status, age, or where you live. They also do not consider salary, occupation, employer, employment history, the interest rate on any account, or the type of credit being requested. Scores are calculated only from the credit report, with one variation: the UltraFICO Score also considers bank account data that the consumer chooses to share. Lenders may still weigh income and other factors in their own decisions, separately from the score.
| In the model | Not in the model |
|---|---|
| Payment history, balances, credit limits | Income, salary, occupation |
| Ages of accounts, number of recent inquiries | Race, sex, religion, national origin, marital status |
| Types of credit accounts | Age, where you live |
| Public-record and collection items as reported to the bureaus | Interest rates charged on existing accounts |
A worked example of utilization
Arithmetic shows why utilization moves a score. Suppose a consumer has two credit cards with limits of $5,000 each, for $10,000 in total available credit, and carries $2,000 in balances. Utilization is 20 percent ($2,000 divided by $10,000). If that consumer closes one of the cards and the balances stay the same, available credit drops to $5,000 and utilization jumps to 40 percent ($2,000 divided by $5,000), even though nothing about spending or repayment changed. The score model sees a borrower who is now using a larger share of their available credit, a pattern that has historically been associated with higher default rates.
Why does closing an old credit card sometimes lower a credit score?
Closing an older account can shorten the average age of a consumer's credit history and reduce total available credit, which can raise the utilization percentage on remaining accounts even if actual spending has not changed. Both effects can push a score down. This is a documented mechanical interaction between two of the five weighted categories, not a penalty for the act of closing an account itself.
How do inquiries affect a score?
A hard inquiry is recorded when a lender checks your credit as part of an application. myFICO says FICO Scores consider only inquiries from the last 12 months, though inquiries stay on the report for two years. Newer FICO versions also treat several inquiries for the same type of loan within a short shopping window (the length depends on the scoring version) as one, so comparing offers from several lenders is not penalized as heavily as opening many unrelated accounts.
Why this is a statistical model, not a judgment of character
Credit scoring models are built with the same broad class of techniques used in other predictive statistics. Analysts take historical data on borrowers, including their reported credit behaviors and what happened to their loans later, and look for which patterns correlate most strongly with future default. The resulting score is a probability estimate compressed into a single number for convenience, not a moral evaluation of the person it describes. For readers who want the statistical background, How Correlation Differs From Causation explains why a pattern that predicts default does not mean the pattern causes it, and How Decision Trees Split Data shows a related family of methods for sorting people into risk groups.
The Federal Reserve Board studied this question in a 2007 report to Congress. It built its own "base model," designed to follow general industry practice, on a sample of 301,536 anonymous individuals' credit records from TransUnion, linked to demographic data from the Social Security Administration. Its core finding was that credit history scores predict credit risk for the population overall and for all major demographic groups: within every group, serious delinquency rates fell as scores rose (Federal Reserve Board, Report to the Congress on Credit Scoring, executive summary).
This framing explains counterintuitive outcomes. A young borrower with a short credit history, even one who has never missed a payment, typically has a lower score than an older borrower with an equally clean record and many more years of demonstrated repayment. The length-of-history factor reflects that a longer track record gives the model more statistical confidence, not that a shorter history reflects poorly on the borrower's judgment. It also explains why two people with identical incomes and bank balances can have very different scores: the model does not see income or savings at all.
How a scorecard turns odds into points
Most scoring models start with odds, not points. Analysts fit a statistical model, often a logistic regression (a method that estimates the probability of a yes-or-no outcome such as "defaults within two years"), and the model's output is a log-odds figure that nobody outside a statistics department finds intuitive. Developers then rescale it with two choices: the score that should correspond to a chosen good-to-bad odds ratio, and the number of points that should double those odds.
SAS, a statistical software vendor, documents a typical example in its scorecard tutorial: a total score of 600 corresponds to good-to-bad odds of 30 to 1, and an increase of 20 points doubles the odds (SAS documentation, Create a Scorecard with a Logistic Regression Model). That is a generic teaching example, not FICO's formula, which is proprietary and differs by model version. It still shows the logic of the scale. Applying the doubling rule gives the table below, where the share of "bad" outcomes is 1 divided by (1 plus the odds).
| Score (example scale) | Good-to-bad odds | Implied share going bad |
|---|---|---|
| 560 | 7.5 to 1 | about 11.8% |
| 580 | 15 to 1 | about 6.3% |
| 600 | 30 to 1 | about 3.2% |
| 620 | 60 to 1 | about 1.6% |
| 640 | 120 to 1 | about 0.8% |
Each 20-point step halves the expected bad rate, roughly. Think of a staircase where every step down cuts the risk in half, so the vertical distance between a 560 and a 640 is a fourteen-fold difference in expected defaults. This is why lenders care about a gap that looks small on paper, and why a score is better read as a ranking of risk than as a measure of quantity. On this example scale, five points shifts the odds by about a fifth, while fifty points shifts them nearly sixfold.
The blind spot: who gets rejected
Models learn from outcomes, and lenders only observe outcomes for people they approved. Someone who was denied never gets a chance to repay or default, so their result is missing from the data. Statisticians call this a selection problem, and it is a cousin of the issue described in How Survivorship Bias Hides Failures. Scoring developers use various methods to correct for it. The practical consequence is that models are most reliable for the kinds of borrowers lenders have historically approved and least reliable for those they have not.
Models are updated, but the structure persists
FICO has released many generations of its scoring model. FICO Scores 2, 3, 4, 5, 8, 9, 10, and 10 T are listed on myFICO, along with auto and bankcard versions. Base FICO Scores run from 300 to 850, while industry-specific versions range from 250 to 900. FICO 9 treats paid-off collection accounts, including medical ones, as no longer harming the score and reduces the weight of unpaid medical collections compared with other debt. It also includes rental payment history when that is reported. FICO 10 T goes further and uses trended data, which looks at balances and credit limits over the previous 24 months or longer rather than only the most recent month (myFICO, FICO Score versions).
The trended-data change illustrates how models evolve. A borrower who pays a card to zero every month and a borrower who carries the same balance from month to month can look identical in a single snapshot. A model that sees 24 months of history can tell them apart. Not every lender uses the newest versions, and the version in use often depends on the lender, the product, and, for mortgages, on rules set by the agencies that buy loans.
Different scoring models exist and don't always agree
FICO is the most widely used scoring brand in U.S. lending, but it is not the only one. VantageScore, a competing model, was created in 2006 by the three major credit bureaus. It was built partly in response to demand for a model that could score more consumers consistently across all three bureaus using a single shared algorithm, an approach VantageScore describes as "tri-bureau," in contrast to FICO's practice of maintaining bureau-specific versions.
VantageScore uses a different underlying methodology than FICO and can produce a different score for the same consumer, even though both typically report on a similar 300 to 850 scale. The two organizations do not disclose identical formulas, and they may treat certain data points, such as recently opened accounts or thin credit files, differently enough to produce a meaningfully different number for some consumers.
A policy change in 2025 shows why the choice of model matters beyond consumers. On July 8, 2025, Federal Housing Finance Agency Director William Pulte announced that Fannie Mae and Freddie Mac, the government-sponsored enterprises that buy a large share of U.S. home mortgages, would accept VantageScore 4.0, with the requirement for reports from all three bureaus left in place. Before then, the two enterprises had accepted only FICO scores, and the announcement was framed as enforcing the 2018 Credit Score Competition Act (MortgagePoint, July 8, 2025). Lender-level rules have continued to evolve since, so anyone applying for a mortgage should check which scores their lender uses.
| Score range | General risk category (FICO scale) |
|---|---|
| 800-850 | Exceptional |
| 740-799 | Very good |
| 670-739 | Good |
| 580-669 | Fair |
| 300-579 | Poor |
This is FICO's general-purpose risk categorization; individual lenders set their own approval thresholds and are not required to follow these labels.
For a sense of where people sit on this scale, FICO reported in April 2025 that the national average U.S. FICO Score was 715, one point lower than in January 2025 and two points lower than in April 2024. FICO tied part of the decline to the return of federal student loan delinquency reporting (FICO press release, April 16, 2025).
| Feature | FICO | VantageScore |
|---|---|---|
| Introduced | 1989 | 2006 |
| Typical range | 300-850 (industry-specific 250-900) | 300-850 |
| Origin | Fair Isaac Corporation | Created by the three major credit bureaus |
| Bureau approach | Bureau-specific versions | Single shared algorithm across all three bureaus |
| Formula | Not fully public; category weights published | Not fully public |
Lenders also sometimes use industry-specific score versions tailored for auto loans or credit cards, which weight the same underlying data differently depending on which type of default a given lender cares most about predicting. A mortgage lender underwriting a 30-year loan and an auto lender underwriting a five-year loan are not necessarily worried about the same risk pattern, so those versions exist to reflect the difference. For a parallel case of an industry pricing risk with statistical models, see AI in Insurance: How Neural Networks Price Your Policy.
Do FICO and VantageScore always produce the same score for the same person?
No. Because the two models use different methodologies and sometimes weight or interpret the same underlying report data differently, they can generate different numbers for the same consumer even when both use a similar 300 to 850 scale. A lender's choice of which model and which version to use is not standardized across the industry.
Why the score exists at all, from a lender's perspective
Before standardized credit scoring became widespread, lenders relied more heavily on manual underwriting. This was slower, harder to scale, and showed documented inconsistencies, including patterns that fair-lending laws such as the Equal Credit Opportunity Act of 1974 were later enacted to address. A standardized statistical score gave lenders a faster, more consistent way to estimate default risk across large volumes of applications. It also gave regulators a documented, auditable methodology to examine, rather than an individual underwriter's undocumented judgment calls.
The Federal Reserve's 2007 report credited scoring with speed, lower cost, standardization, and more consistent decisions, which may reduce the influence of prohibited factors, and with enabling pricing that better matches risk. It cautioned that direct evidence on scoring's effect on credit availability is limited because the technology became widespread decades ago.
Lenders use the score primarily to answer one question: given everything the model can statistically infer from this credit file, how likely is this borrower to repay as agreed? A higher score corresponds to a lower estimated risk of default, which is why lenders commonly offer more favorable terms to higher-scoring applicants. The same logic links scores to broader rates and credit conditions, a theme covered in Why Interest Rates Affect Everything. The Fed's study found that loan pricing and inferred denial rates both decline as scores increase.
The score does not guarantee any individual outcome. It is a probability estimate applied to an individual case, and probability estimates are, by definition, sometimes wrong for any single person even when they are accurate on average across a large population.
What credit scoring leaves out, and documented criticism of the system
Credit scoring models depend on having enough data in a consumer's credit file to analyze. Consumers who have never used traditional credit products, such as recent immigrants, young adults, or people who primarily use cash, may have too little data for a model to produce a reliable score. The Consumer Financial Protection Bureau (CFPB) published a widely cited estimate in 2015 that about 11 percent of U.S. adults (25.9 million) were "credit invisible," meaning they had no credit record at the nationwide bureaus. In June 2025 the bureau published a technical correction using a newer data source and concluded that the original estimate "should be roughly cut in half," with a nearly matching rise in records that exist but are unscored. The revised 2010 figure is 5.8 percent of adults (13.5 million of 235 million). For December 2020, the bureau estimated that 2.7 percent of adults (roughly 7.0 million of 258 million) were credit invisible, 9.8 percent had an unscored record (5.9 percent stale and 3.9 percent with insufficient information), and 87.5 percent had a scored record (CFPB, Technical correction and update to the CFPB's credit invisibles estimate, June 2025; original: CFPB, Data Point: Credit Invisibles, May 2015).
| Group (share of U.S. adults) | 2010 as first reported (2015 report) | 2010 after the 2025 correction | 2020 (2025 report) |
|---|---|---|---|
| Scored credit record | 80.7% | 81.6% | 87.5% |
| Unscored record (stale or insufficient) | 8.3% | 12.7% | 9.8% |
| No credit record (credit invisible) | 11.0% | 5.8% | 2.7% |
Source: CFPB, June 2025, Tables 1 and 2. Columns may not sum to 100 percent because of rounding.
The table carries a lesson about measurement itself. The people who were "missing" in 2015 had not vanished. A better data set simply showed that many of them had records containing only items such as deferred student loans, collections, or closed accounts, which a scoring model cannot turn into a score. The share of adults who can be scored rose by almost six percentage points over the decade, a shift the bureau noted is broadly consistent with a fall in the unbanked household rate from 8.2 percent in 2011 to 5.4 percent in 2019 (FDIC survey data cited in the CFPB report).
The Federal Reserve's 2007 study documented that scores differ across demographic groups. Black and Hispanic consumers, single individuals, people under 30, and residents of low-income or predominantly minority census tracts had lower average scores, and controls narrowed the gaps without always eliminating them. The study also found that performance relative to score varied: Black consumers performed somewhat worse than other groups with similar scores, while Asians, married individuals, and recent immigrants performed better than their scores predicted. When the researchers re-estimated their model in a race-neutral setting, scores changed little, which suggested the credit characteristics were not acting as proxies for race or ethnicity. Length of credit history, however, partly proxies for age, lowering scores for younger people and recent immigrants and raising them for older people. The report suggested age-neutral weights as an alternative to dropping the characteristic, since it has strong predictive value. The authors noted limits: they lacked data on wealth, employment, and education, and credit files contain little demographic information, so some differences could not be fully explained (Federal Reserve Board, 2007).
A newer line of scrutiny concerns machine-learning scoring approaches that use "alternative data," meaning information beyond a traditional credit file, such as utility payments, rental history, or behavioral data. Federal regulators have raised concerns that some alternative-data models could produce discriminatory outcomes, even unintentionally, if the underlying data reflect historical inequities or include information not tied to an individual's own repayment conduct. The same debate shows up in AI Ethics: The Key Debates. These are documented, ongoing regulatory and research concerns rather than settled conclusions, and any biases embedded in historical data or in the choice of data sources can, in principle, be reflected in resulting scores.
Medical debt is one visible example of the policy tug of war. FICO 9 reduces the weight of medical collections, but whether lenders use that version varies. A CFPB rule that would have removed an estimated $49 billion in medical debt from the credit reports of about 15 million Americans was vacated by a federal court in Texas in July 2025 on the ground that the rule exceeded the agency's legal authority (Consumer Finance Monitor, July 15, 2025). The outcome illustrates that what counts as predictive and what is permissible to include are two separate questions.
How long old mistakes stay on the record
A score can only use what the report contains, so the reporting clocks matter. The FTC says most negative information stays on a credit report for seven years and bankruptcy information for 10 years (FTC, Fixing Your Credit FAQs). The CFPB adds that credit reporting companies may report positive information for longer, so an account paid on time can keep appearing after it is closed, and that the seven-year limit does not apply to reports used for a job paying more than $75,000 a year or for credit or life insurance above $150,000 (CFPB, How long does information stay on my credit report?).
The practical effect is that the model's picture of a borrower is always a moving window. A missed payment that sat in the "recent and severe" bucket last year drifts toward "old" with each passing month, and most scoring models give recent events more weight than old ones. This is the statistical reason a score can recover without any dramatic action by the borrower: the same record, viewed from a later date, predicts less.
The FTC also warns that credit repair companies cannot remove negative information that is accurate and current, and that anything such a company can legally do, consumers can do themselves for little or no cost. That matches the logic of the model. If an item is true and within the reporting window, the scoring model is correctly describing the past, and the only legitimate fixes are accurate reporting and the passage of time.
Why lenders must explain a denial
A model that nobody can explain is hard to audit. Under the Equal Credit Opportunity Act and its implementing rule, Regulation B, a lender that turns down an application must tell the applicant why. In a May 26, 2022 circular, the CFPB stated that this duty applies whatever technology the lender uses, that the statement of reasons "must be specific and indicate the principal reason(s) for the adverse action," and that citing a failed score or internal policy is not enough (CFPB Consumer Financial Protection Circular 2022-03). The circular also says a lender's inability to understand its own model is not a defense.
This requirement shapes how scores are built. A traditional scorecard assigns points to each characteristic, so the biggest point losses can be listed as reasons ("balances too high relative to limits," "recent late payment"). More complex machine-learning models fit patterns that are harder to translate into plain reasons, which is one reason regulators scrutinize them. Readers interested in the modeling side can compare the transparent additive structure above with the tree-based methods in How Random Forests Reduce Error, which trade some interpretability for flexibility. The circular reflects the CFPB's reading as of 2022; the binding requirement is in Regulation B itself.
Common misunderstandings
| Belief | What the evidence says |
|---|---|
| "My income affects my score" | FICO Scores do not consider salary, occupation, or employer. Lenders may look at income separately. |
| "Checking my own score lowers it" | Checking your own score is not a hard inquiry. Hard inquiries come from applications for credit. |
| "Closing a card always helps" | It can raise utilization and shorten average account age, both of which can push a score down. |
| "There is one credit score" | Consumers have many scores across FICO and VantageScore versions, and lenders choose which to use. |
| "A high score means I am wealthy" | A score reflects borrowing and repayment behavior only. It says nothing about savings or net worth. |
| "A low score reflects bad character" | The score is a statistical estimate based on credit-file patterns, not a moral judgment. |
What rights do consumers have regarding their credit report?
Under the federal Fair Credit Reporting Act (FCRA), consumers can request a free copy of their credit report from each of the three major bureaus through AnnualCreditReport.com, and can dispute information they believe is inaccurate. The FTC says the bureau has 30 days to investigate a dispute, must forward the evidence to the business that reported the information, and must give the consumer the results in writing. The FTC also notes that the bureaus have permanently extended free weekly reports through the same site (FTC, Disputing Errors on Your Credit Reports). Consumers should direct disputes through the bureaus and the reporting business, which are the legally established channels.
For how scores fit with day-to-day money management, see How to Build an Emergency Fund and How to Create a Budget Using the 50/30/20 Rule. Those articles are educational overviews and not personal advice.
The takeaway
A credit score works by converting patterns in historical repayment data into a single number that estimates the statistical likelihood of future default, weighted most heavily toward payment history and credit utilization because those two factors have shown the strongest relationship with default risk in large datasets. Understanding the score as a forecast rather than a verdict helps explain why specific mechanical changes, such as a jump in utilization after closing an account, move the number in predictable directions, and why the system draws sustained scrutiny over data gaps and bias. The score is also less singular than it appears: many models, versions, and lender policies sit behind the one number a consumer sees. This article is educational and does not constitute financial advice; questions about an individual credit file or a lending decision should go through the credit bureaus, a lender's official dispute process, or a qualified financial or legal professional.
FAQ
Q: What is the range of a FICO credit score? A: FICO base scores range from 300 to 850, with higher scores indicating a lower estimated risk of default. Industry-specific versions, such as auto and bankcard scores, range from 250 to 900. FICO introduced its general-purpose score in 1989.
Q: What factors make up the largest portion of a FICO score? A: According to myFICO's published methodology, payment history accounts for about 35 percent and amounts owed, including credit utilization, for about 30 percent. Together they make up roughly 65 percent of a typical FICO Score. Length of history is 15 percent, and new credit and credit mix are 10 percent each.
Q: Does a credit score consider my income? A: No. myFICO states that FICO Scores do not consider salary, occupation, employer, or employment history, nor race, sex, marital status, age, or where you live. Scores are calculated from the credit report. Lenders may consider income separately when deciding whether to approve a loan.
Q: Why might closing an old credit card lower a credit score? A: Closing an older account can shorten average credit history length and reduce total available credit, which can raise the utilization percentage on remaining accounts. Both factors can push a score down even without any change in spending behavior.
Q: Do FICO and VantageScore always produce the same credit score? A: No. They use different methodologies, so they can produce different scores for the same person even though both typically use a 300 to 850 range. Lenders may also use industry-specific versions, and until July 2025 Fannie Mae and Freddie Mac accepted only FICO scores for the mortgages they bought.
Q: What does it mean to be "credit invisible"? A: Credit invisible describes a consumer with no credit record at any of the three major bureaus, so there is nothing for a model to score. The CFPB's 2015 estimate of 25.9 million adults (11 percent) was cut roughly in half in a June 2025 correction. The bureau now estimates 2.7 percent of adults (about 7.0 million) were credit invisible in December 2020, with another 9.8 percent holding records too stale or thin to score.
Q: Can I dispute an error on my credit report? A: Yes. Under the Fair Credit Reporting Act, consumers can request free reports through AnnualCreditReport.com and dispute errors with the bureau and with the business that reported the information. According to the FTC, the bureau has 30 days to investigate and must report results in writing.
Sources
- Federal Reserve Board, Report to the Congress on Credit Scoring and Its Effects on the Availability and Affordability of Credit, August 2007.
- myFICO, What's in my FICO Scores and FICO Score versions.
- Federal Trade Commission, Disputing Errors on Your Credit Reports.
- Consumer Financial Protection Bureau, Data Point: Credit Invisibles (2015) and Technical correction and update to the CFPB's credit invisibles estimate (June 2025).
- Consumer Financial Protection Bureau, How long does information stay on my credit report? and Circular 2022-03 on adverse action notices.
- Federal Trade Commission, Fixing Your Credit FAQs.
- SAS, Create a Scorecard with a Logistic Regression Model.
- MortgagePoint, Fannie/Freddie Approved to Use VantageScore 4.0, July 8, 2025.
Read next:
Follow new articles
Add SciHub101 to your feed reader to see new science and technology stories.
About the Author
SciHub101 is an independent science and technology publication founded and edited by Buğra Sözeri, a software engineer and app developer based in Türkiye. AI tools assist with drafting and research. Source documentation in the existing archive is uneven, and articles are being reviewed and corrected. The Editorial Team byline identifies editorial responsibility; it does not certify specialist or independent human review. Where we compare products, anything built by our own team is labelled as such — see the editorial guidelines and About page for how corrections and disclosures work.
Related Articles
Complete Finance Guide 2026 — How Money Really Works
The complete guide to finance in 2026: how money, inflation, and central banks work, investing fundamentals, markets, cryptocurrency, and personal finance explained.
