Data Analyst Portfolio Project with AI | NBA Championship Prediction Model (End-to-End Project)

Okay, so the new NBA season is about to tip off and we are going to build a data analyst portfolio project predicting who will win the championship this season with the help of Claude. Now, a few months back, I made a video where I did the same thing predicting the 2026 FIFA World Cup, and my model ended up being correct with Spain as the winner. Now, everyone already has their championship pick. Your group chat has one, the guys on TV have one, And I've got mine too. I want the Knicks to win it all again, but today data makes the call, not me. We're gonna be doing this completely end to end. We'll fetch the dataset, we'll build the project requirements, clean the data and do the analysis, and finish with an actual portfolio one-pager that you can hand to a hiring manager. Now, if this is your first time here, I'm Kedisha, a self-taught data scientist and instructor turned data career coach. And for over 5 years, I've helped thousands of other career changers break into data analytics without going back to school or wasting more money on courses that just don't work. And if you ever want to work with me to land your first role in data, click the link in the description. Now let's get started. Okay. So last time when we did this, I used Fable because it was brand new. And to have the same type of, I guess, prediction effort, we're going to use Fable again. Looks like there's different efforts. You got low, medium, high, extra, and then to the max. We're just going to keep it at medium for now. Okay. So the thing with this project is we want to make sure that Claude's not going to pull from Kaggle. And we also want to make sure that it doesn't like go ahead of us to like start cleaning the data or, you know, analyzing things that we don't want it to yet. This is going to be an exercise of directing AI basically, right? Okay, so I'm going to enter in, hey, I want you to help me create, and I have my old prompt for the previous video. So I'm going to peek at it just to see how I remember how I started it. So yeah, I want you to help me create a data analytics portfolio project. Basically, an NBA championship or champion prediction. I want you to source raw data, create a project plan. With a business problem visualization with tables and start from there. I do not want anything from Kaggle, real data only. Do not start cleaning Or analyzing anything until I say so. Um, our goal is to beat whatever the current Vegas odds are. Okay. Yeah, so the MA season, uh, preseason has been going on, but The, uh, see, actually regular season has not started yet, and our goal is to basically beat, um, whatever the current Vegas odds are. Okay, looks like we're going to use the current DraftKings team odds. Okay, so created a scenario. We are an analyst for a sports analytics consulting pitching a hedge fund style sports investor. The investor doesn't want a hot take on who wins. They want a repeatable data-backed probability for every team, a comparison against the market, and a short list of positive expected value positions. The question that our project's going to answer, which teams does the DraftKings Futures board Over or underpriced for the 2027 title and by how much? Okay, so it looks like we have the odds currently of what DraftKings have. So they currently have basically the Spurs, the San Antonio Spurs and Oklahoma City Thunder as the tied to have the most, the highest probability to win the championship. The Knicks are the current defending champions and they are Not the favorites. So just thinking of what goes into how to build the best model to predict the winner. That's basically what we're doing here today. How do you, how do you like create the best model? And what goes into models that go, you know, data. What's the best? If we think about like the best type of variables. Or predictors for a championship team. Now looks like they we got a project plan ready, so this is good. Right? Draft Kings have the Spurs and Thunder as co-favorites. These are our data sources. Looks like it's like NBA Stats API championship listing from. 2010, though I don't think, I don't think it really matters. So this is something that we're gonna like, these are gonna be our data sources. When it comes down to the predictors of the championship, I honestly don't think, you know, data from 2010 to that early, like that beyond, you know, 16 years ago matters. Like what a team was 16 years ago has no predictor of what they'll do right now. So again, so this is going to be an exercise on directing AI. So I already know that when it comes down to historical data, I don't think data from 2010 matters now in terms of the championship. But unlike predicting, say, like a FIFA World Cup where it's every 4 years, It is the best of the best representing each country. And it's a very— it's a much shorter, confined amount of time. So to me, there's less variables there. Talent in terms of like roster and like individual talent matters a lot in sports and team sports. But again, the FIFA World Cup is like, it's just a summer. But when it comes down to an entire season, so many things can be— so many variables. So. roster, individual talent. So for example, the Knicks won last season, but they're obviously not the favorites, predicted favorites. And even though that's my favorite team, I, statistically, we're going to create the best model to, again, have the best, to find the best odds. But thinking about the predictors there. So it's an 82-game— it's 82 games in a season, right? And so when it comes down to basketball, especially in the past decade, injuries has been huge, right? So the 76ers, for example, this is a team that the makeup of the team is completely different from a year ago, right? They've made a lot of roster moves. A lot of teams over the past— over this Summer has made a lot of roster moves, but they have on paper they look really good, but they have some injury-prone players, right? So the probability of injury is something that I want to make sure we have in our data, right? So when I look at the datasets that Kaggle has provided, we have our benchmark to beat in DraftKings. Nice. For one thing, I already know for championship list, I don't really, I don't want to go so far back, to be honest. But when I see the stats API, I'm seeing team roster, agent experience, per player minutes, age usage, traditional, but I don't see anything with injuries. So I'm going to tell Claude, hey, for the Data sources, I believe we should include you know some data slash variable for you know odds of injury. Maybe for a player, to be honest with you, it could be we could add something where it's like injury. You know, how many games someone will play, maybe? Yeah, so, like, some sort of, like, data variable for odds of injury or games played. For example, Joel Embiid has a large amount of injury history. We also want to take into account of, like, postseason in the NBA. For whatever reason, some players do very well in the regular season and they are— they completely suck in the postseason, in the playoffs. Also, you know, and, you know, also something regarding, you know, players performance in regular season versus postseason. Like, yeah, um, and for predicting the champion, I honestly do not believe historical winners, you know, backed all the way to 2010 means anything since, you know, NBA rosters change so often. All right, so I'm gonna put that in there. So like right now what we're doing is Trying to define or get like the best data that we want to do. So again, when it comes down to a model, it's garbage in, garbage out. Like the data that we choose and the variables that we choose in this beginning matters, pretty much defines everything, right? So again, I don't see anything with injuries. I think injury is going to be a big predictor. You know, when it comes down to like team statistics, for example. Again, NBA rosters change all the time. And this season, just because basketball's my favorite sport, I know, you know, in terms of like extra information, like there was so many roster moves over this summer. So a team's defensive, you know, if a team was like, some teams like, you know, maybe they were great defensively, Mm-hmm. last season, but then their best defender changed teams or something like that. So that's going to affect. So that's why we— OK, good. So we updated it. So yeah, as Claude says, I'm right on all 3, and the 3rd change is the model design. So they added injury availability, 2 layers, games played per season, and then plus we're going to include an injury database. Great. And we're going to have an availability. We're going to build an availability score per player. So that's cool. And then a team's projected strength gets discounted by it. Excellent. Regular season versus postseason. Also, yes. Right. We're going to have, you know, stats by season, by type. So a playoff minus whatever it is. So that's good. Right. Some, some people, they're great during the regular season and they're complete opposite during the postseason. All right. So Claude wants to push back on the historical winners. You're— I'm right that the franchise history means nothing, right? 2016 Cavs tell you zero about the 2026 Cavs. So the unit of the model shouldn't be the same season. It should be the player. So you project each player on the current roster from his own last 3 seasons. Excellent. That I like. But you still need past seasons for one thing. yardstick. Something has to tell you a team with strength X wins the title about Y all of the time. And the only way to learn that is from seasons that's already happened. Okay. Okay. So now let's see one thing you'll do by hand, the pro sports transactions history report. It's a search form, not an API. Pro sports injury log. They're saying that this is so okay. Oh darn. All right, all right, all right. Okay, so injuries is going to be a little harder because it is searched. Oh my gosh, darn. Okay, so let's not do injury type for injuries. So it will be manual work for the— for the injury report is going to be out of scope because there's no way I'm going to be able to just do player by player. It's just not gonna be possible. So it's gonna be out of scope. What's the alternative? What's the automated alternative? Alright, so we need to figure out how to include a variable for player availability without Having to manually pull through their injuries. Let's see, can Claude scrape it instead? Otherwise, we can just do player. Okay, looks like Claude's going to try to scrape. I need this. Okay, looks looks like they pulled the injuries themselves. So looks like Claude scraped and pulled the injuries. So let's do. If you want automated, let's see, to hand entered. I need, I need everything automated. I'm not doing anything by hand. Okay, so right now we're still just trying to finalize our initial data. Again, the initial data is everything. The initial data is everything to make sure we have the best, to make the best model. That we want. Okay, so let's look at where we're at now. So what we wanna do is we wanted to update the data so that, again, we incorporate player availability because players, you know, they get injured, etc. So, like, while, you know, a team can look great on paper, If someone has, you know, historically they get injured a lot or whatever it is, if they're not available, then, you know, that just lowers the roster. You know, that just lowers the— that should, you know, that lowers the chances a team will do well, you know, if their better players are not playing per se. We're also going to factor in Players who, they're very good in the regular season, but they're not very good in the postseason. In terms of data going back to 2015, the idea is just to see, again, more so about the individual player, their history of success and their history, whatever, rather than the team, et cetera, per se. Right? Because, you know, the Knicks in 2015 does not have anything to do with the Knicks' success in 2025. Okay, so let's move on to— okay, I like the data we have. Let's move on to the next step. Okay, this has to happen on my machine. All right, let's see what else is on the internet that we can actually use. Okay, so update. date. I went ahead and had Claude just provide me those initial local scripts to run to pull all the data that we need. I had tried to say, oh no, just pull everything, just find stuff on the internet, but it was taking like 25 minutes and still wasn't like pulling anything. So I decided just like, okay, I'll just provide me all the scripts. It has about 3 different ones, and all you have to do is run them locally on your machine, on your command line. So there's a pull raw data, pull injuries, historical odds, so that we can have all the data that we need for step 1. So this is like the README here. So again, we have— these are all the data points that we're going to use to build our model. And the main thing is We are looking at historical data from 10 years back, and we want to basically not necessarily focus too much on the team historical performance, but the players' performance, their players' history for the past 10 years, right? So we're going to get their injuries, how well they've played in the postseason versus the regular season. And all that stuff like that so that we can have like a great picture to build a great model or the best type of model incorporates, you know, multiple things that we need. Additional thing that we wanted Claude to do was we wanted to have data on injuries so that we can create an availability score, which will pull again injury logs and also like how many games people played. Right. Sometimes people play less games because of multiple things. You know, sometimes it's suspension, for example. But like, if someone has like a history of like someone did like an Achilles injury or something like that, and, you know, that's more severe than maybe like, you know, sprained their pinky toe or something. So, so that is what I went ahead and did. And what we ended up getting was 114 files. And I will actually have all this data. I'll share the data so you can also do this yourself. So if you don't want to run, use a command line to run to get all the data, you don't have to, but I'll have that there. And so what I did was, you know, Claude has like, you can only upload 20 files at once. And since we have 114, I just uploaded everything into a Google Drive. And at the moment, the agent's just running through. I will say this has— the data gathering process has definitely gone longer than the FIFA one. If I have to think about it, I probably, to be honest, I think— You know what? Let me Pause this because I'm like, rather than go back 10 seasons, and then this will also help with the— like, let's just go back. Let's just limit the data to the previous— again, instead of 10 seasons, you know, yes, there's some players that are playing Now, and they were playing 10 seasons ago. But, you know, I don't think what they were doing again 10 seasons ago matters as much. So I think honestly, just like with the FIFA one, rather than going back that far, and this will also help save time with the data, I say just the previous— Um. We can just limit it to the previous, like, I say limit the data to the previous, uh, 3 seasons. I don't, I really don't think it matters to go that far. Let's just limit the data to the previous 3 seasons. And this should also, um, this will, uh, 3 seasons it is. Yeah. So it's also gonna help save time. Should have just thought about that before. That's also one thing that Claude, Claude, actually for the FIFA one, they wanted they they recommended going back like fifty years, which I'm wasn't fifty plus years, and then I'm like why do we have to go back that far? Okay, I already have the data. I already have the data. It's in this. Just pick out. Just pick the lines where. It's limited. Attach one zip here. All right, let's just, uh, run this on our machine. All right, they just put it somewhere else. All right, so now we have our new zip file. Okay, so what we did was we were just filtering down the data to the last 3 seasons, and we asked Claude to Yeah, all we did was just filter it down to the last 3 seasons. Okay. Audit data is in good shape. Oh, decisions before cleaning. All 3 seasons checked out. We got our playoff winners, Celtics, Thunder, the Knicks. Okay. What needs cleaning? Traded players get merged. NBA.com rolls a midseason trade into one row. Someone shows 84 games because he played for 2 teams with different schedules. 19% of the 2026-2027 rosters have no stats. Well, yeah, because they're rookies. So treat them as replacement level and flag the handful of first-round rookies separately so Cooper isn't priced like an end-of-bench guy. That's true. Stats are season totals, not per game. This second point. So there's really— that's one thing too. There's really no Predictability when it comes down to rookies. Let's say second-year players, you know, players that don't have as much history. Let's see, they don't have as much data as like players that's been, you know, playing that played all three previous seasons, right? Yeah, we'll just do this. Treat them as just do this. You know, we don't want to approach. But the idea is that, you know, Cooper, you know, a player like Cooper Flagg. So like, let's say a rookie who had a very good rookie season, he only has 1 season of data versus maybe like a scrub with 3 seasons, you know. A player like Cooper Flagg with only 1 season of data should not be seen worse than a scrub who's played all 3 seasons. Now I will say this now gives me an idea for additional data to— for this, for the rookies. So we don't have any data for them. So what is a predictor of success maybe for rookies? Easiest thing I can think of— I don't want to pull— I don't want to use college data. Or if people played overseas, I say we pull, we just add like a very plain, we just use their draft, right? So something where like, hey, a player who was drafted high, they'll have maybe a higher predictive, they'll probably make some sort of higher, just for the rookies, right? We can use like where they were drafted. So like player that got picked first, he has like a higher predictability score of whatever it is. And then if you're you were drafted last, and then if you're undrafted as like a flat standard score. So let's do something like that for rookies. Let's use a per let's you know make a score based on where they were drafted. Anyone who Was undrafted gets a standard score. Let's do that. That's something that's that's something that could just be pulled online. So there's there should not be that should not take long. And then I think once we have this, yeah. So rookies scored by draft slot undrafted a standard score. So no one gets penalized for having fewer seasons of data. Perfect. It also makes it clear, had we gone the alt 10 seasons, right? Um, I mean, that's something I didn't even think about. Um, that would have been a huge gap there. One gap: 22 rookies came in via draft rights trades, so their pick number isn't on the roster file. Oh yeah, the 2026 draft board fills that in. Okay, perfect. Okay, I think once we have this, then we should be good to go to Let's see, what's the next step? So basically what we're doing now is that, you know, we have rookies that have no data. And what is the idea of like, what's the best predictor of how they would do? We can use, we can use like where they were drafted. So that's like the really the only thing we can use. And I don't want to use college data because I don't think that— I mean, that would be— make it— that would make it way more complicated. That would make it way more complicated. And then there's just way too much variables there. You know, some people were on like way better teams, way better talent. So maybe their stats, you know, the stats, you know, not all stats are going to be equal if you're going to incorporate college. Some players were on less talented teams and they probably, you know, had a whole bunch more shot opportunities, etc. So this is going to be a good means for addressing the rookies that have literally no data. Okay. Paste there is done. We have 3 clean tables: player season CSV, we have regular season rosters. Okay, go to the next. Let's go to EDA now, exploratory data analysis. Okay, next step now. Okay, we ran an availabilities. All right, 3-year availability of each team's top 8 rotational players. So that's key, the top player or the second best player getting injured affects the probability of a team winning versus a scrub getting injured. So looks like in terms of the availability score, and this is going to, this uses data for the history of what the games people played, how many games they played, also injury. So San Antonio Spurs, like I said, they're a very healthy young team. So they're number 1. The New York Knicks are number 2, defending champions. So The Knicks and San Antonio, they were the top 2 teams. They were the, they were in the finals. So if we remember what Vegas has, they have 4 teams basically tied for 1 and 2. San Antonio and Oklahoma City Thunder, top 2 for, well, top 2 in general. But when it comes down to the NBA, it's, you know, 2 conferences. So you have to, the teams that win each conference, West and East. Though they meet in the finals. So the East were New York and Philadelphia. And when it comes down to the West, it's San Antonio and OKC. So let's see who's third. The Hornets, but they're not going to go anywhere. And then Atlanta, but you know, they're not really— okay. So, but so these are going to be the top teams just on availability score. And again, these are based off of like individual players. So. We see Philadelphia. So Philadelphia is in the middle. So they have a 74% probability of the top 8 rotational players. So that's a big deal. So interesting, interesting. And let's see if there's any other flags here. Miami. Miami is a team that's done a lot of changes too. That's gonna have higher expectations and theirs is a 74%. Okay. The 2 co-favorites sit at the opposite ends of the charts. Thunder's in the middle of the pack. Yep. They're one of their top players home. Yep. That's, this is their starting center, starting power forward. Holmgren, Chet Holmgren. Yeah. He's had a lot of injuries. He's had injuries, like very good player, but again, If you're not available, you're not available. All right. So yeah, so we have definitely outliers here. Let's see, playoff risers and failures. League-wide baseline matters. Roster value the wrong way. Championship champion profile. All 3 champions for the past 3 years had top 5 offense and top 5 defense. All right. All right. Everything is saved. So that's interesting here. So. I mean, interesting. I mean, this makes sense. Availability scores. Okay, let's move on to number 2 because this is going to play into the team strength availability. Because again, you can have the best players, but if they get injured or they're not available, then, you know, that shouldn't— that lowers your chance for our model. All right, let's move on. Let's move on. All right, so we're still doing EDA. And right now we're going to be doing some feature engineering. And like I said, yeah, step 1 was the hardest, really the longest, but the most important. All right, we're going to be building player value score. Every player is going to have a value score and then every team's going to have a strength score. Yep. Rookies are going to have their value based off of where they were drafted. Feature engineering. There's a lot of money in feature engineering, you guys. So it's very good to learn feature engineering. Feature engineering is definitely a 6-figure skill. So all the things that we're talking about in terms of like, oh, hey, we should, you know, we should incorporate some sort of availability score due to like injury history, people who are injury prone. Oh, how do we score our, you know, rookies that have no data? Okay, let's think of using the draft, you know, where they were drafted. So those are, those are strategic things that you as an analyst, data professional, You come up with for the feature engineering, right? The data is the data, but then it's like, okay, how can we engineer features? And also, so again, AI is your copilot. You are the pilot. Knowing what to do is better than knowing exactly what to do. Yeah. Knowing what to do is better than knowing necessarily how to do it. Because again, we knew what we wanted to do and then Claude is basically executing it for us. Okay, let's look at what we have here. Phase 5, we're building our player value score and team strength. So here's just how the score works. Every player has a value score and every team has a strength number. We have impact. Can you show— okay, with 3 seasons, the plan is a regression on the strength of expected net rating without the historical odds. There is no past. Vegas score to score against, so the backtest only checks the model against outcomes, not against the market. Wait, why did we not— why did we remove historical odds? When did we remove historical odds? I only wanted to just go back 3 seasons. Okay. We're gonna fetch some historical odds for the past— Vegas odds for the past 3 seasons, just so we have something to compare against. But again, it's going to be tricky because, you know, say like the Milwaukee Bucks, for example, their best player is now on the Heat, you know, from over the summer. So, you know, Milwaukee's odds from last season than the season prior to that has nothing to do with their odds this season. Like, it's a different team. Their best player is no longer there. So, but we will just take these or scrape some odds just to have some, have some history to backtest. Make a new script to pull the store Make a new script for me to run the historical odds on my machines. This is taking too long. This is Fable, so like, you know, it's just more expensive. I'm not sure how I missed the historical odds, but we're just gonna run it on my machine and then give the data. This is just taking— I don't know why it's taking so long. Wait, do I need to run it? Or do you already have it? Okay. Okay. So I don't need to run the script. You have the data. Okay, perfect. Okay, so let's go. Okay, next phase. Yeah. So again, like the Milwaukee Bucks' odds of winning last season is completely different from this season, because again, the makeup of the team. So Predicting title odds for the NBA, the NFL can be— well, I would say more so in the NBA. There's less fluidity in the roster changes season to season. In the NFL, if you were to do a prediction for the Super Bowl, which I could still probably do, it wouldn't be the exact same just because the season already started. But it's not going to be the same because the season already started and then there's already been injuries. For example, Whatever odds people had for the Giants is different because Dart got injured and stuff like that. I mean, same thing even with the, you know, doing this type of project is much better now. It's different than say like doing it, you know, one month in. It's just not the same. All right. We're going to do a regression from team strength to expected net rating. Then a simulated playoff bracket that turns 30 net ratings into 30 title probabilities that sum to 100, and then gets scored against the 3 market boards. And our market boards are DraftKings, FanDuel, and this consensus. I'm not sure if that's— let me make sure that it's— after this runs, we're going to make sure that the current Um, we got DraftKings title odds, the current ones. Actually, no, I think we already have that. Yeah, it's just doing it historically. Yeah, that's what we started with, with OKC and San Antonio tied number 1, and then Philadelphia and New York tied number 2. And that's really just based off what the data is. Those are like, again, just because it's East and West. They're basically saying that, um, it's gonna be down to those 4 teams and stuff like this. It's like there's some team that's just like, you know, like Memphis, for example. That's, that's, they're just, it's just not happening. Um, so it's really just come down to like a few key teams. There's less Cinderella stories of like, you know, teams that come outta nowhere that you just not expect. Like, there's always like a few teams that's like, yeah, but like rarely is it like. Yeah. Crazy unexpected. Okay, our scorecard. Let's see, the market wins 2 seasons, both on scoring rules. Okay, so it looks like my model. Okay, I'm gonna ask it to like, okay, explain it in plain, explain to me like I am 5. Because I'm like, what are you talking about? Just explain to me like I'm 5, which is great, right? Okay, picture it like this. What we built, we graded every player like a report card, then added up each team's report cards to guess how good the team is. Then we played the playoffs. We did a simulation. Vegas also guesses how each team is, but they use everything— the players, the coach. Oh, I didn't think about that. The players, the coach who got hurt in practice yesterday. Well, technically we don't have Who got hurt in practice yesterday? The test, we went back to test the last 2 seasons, I guess, with our model and made our guesses only using what we would have known before the season. I mean, Vegas, yeah, because Vegas can always update the result. Oh, you know what? I wonder if— hold on, let me ask. Like, I wonder if, like, for example, if like, because we're doing the thing like before the season started, so like— Whatever Vegas had, you know, prior to the season, that's what we should be comparing our model against versus what Vegas had like at the end of the regular season after, you know, like, because we don't have that information. So let me see, are we comparing our model against preseason, preseason predictions or You know, Vegas predictions. Let's see, as the season went along, because technically I do not have that data since the season has not started. All right, just asking. Preseason only. Okay. For the backtest, the model only used data for before the season. Vegas number came from the opening week board before any games were played. All right. Where midseason's odd could come. So the backtest you have is legitimate. The only remaining decision, add continuity or retest. Okay, explain the continuity again, because it looks like the model that we came up with, basically Claude is saying it's not very good. All right, explain the continuity test. Again. Okay, good. How the continuity— basically, we want to add in an ingredient. How good was this team last year? And how much of that team is still there? Okay, so this goes back into what I was saying. For example, like the best player from Milwaukee, right? You know, Giannis is now on the Heat. So yeah, let's do that. And we're also going to incorporate the coaching staff. Okay, good. So let's do that. So now we're going to add in more data, and we're going to add in roster continuity in terms of like— that's basically, we were considering those things. It's just, I guess we didn't have enough data to account for it. And I wonder if we should also incorporate the coach. Like every coach has, like every head coach has their own historical overall win-loss record. Let's see if we can also do that too. All right. Version 2 is in better, but still not as good. Okay. All right. So it's close. The player-based model recovers most of the MERT rankings. Okay. The other, okay. Obviously it's like they're incorporating a lot more data like coaching. Okay. So let's, let's add in, let's add Let's add in coaching, head coaching record. Let's add in head coaching record because the Knicks, for example, we basically had the same roster and we changed the head coach, and that did wonders. We went from Coach Tibbs to Mike Brown. Mike Brown is— I mean, they're both defensive guys, but— Coach Tibbs has always had like a history of overplaying our players. And they, you know, like overplaying way too many, too many minutes for the starters and just like not enough. And then that would like, you know, run down a player's body and things like that. So yeah, so let's, let's add in the coach's track record. There's probably other data and stuff like that that the other models will incorporate that's probably just going to be too complex for us, but let's just, you know, see what we can do. Let's just see what we can do here. We're going to add in data of a head coaching record. All right, it looks like the coaching record, or adding the coach, didn't do much. Yeah, ship without it. Okay. Ship without the coaching record. Okay. So, so far, it looks like the model didn't predict, wasn't correct for the previous couple seasons, but let's just see what it pulls out now. I mean, when it comes down to, I mean, at least for me as a basketball person, like this is purely, I mean, there's so much data that when it comes down to like predicting, you know, sports betting, we're just predicting a winner. There's gonna be objective stuff, subjective stuff, stuff that like, like for example, I'm a huge Knicks fan and I pre— you know, I always predict them to win, but I'm pretty sure they were not, they were not seen as the favorites last season. They still won. So there's a lot, lot, lot, lot of data and variables that's hard to really capture unless you get really, really, really complex, which I'm pretty sure DraftKings and FanDuel and Vegas, they, they have all types of, you know, data scientists and analysts, you know, adding all types of data. They, they probably use, you know, insider reports or whatnot. So we're, we're just gonna run ours and see. What our model predicts and just run with that. You know, this is definitely more complicated than the FIFA one. FIFA one honestly just has a lot less variables just because it's like it's every 4 years and, you know, the— it's the best of the best. Stuff like the NBA is like just, just way more variability and like just so many, so many other things to encounter. So it was a lot easier in my opinion for The FIFA one. Yeah, with FIFA World Cup, like taking the previous few World— definitely the historical, just because it's the best of the best. Same thing with like the Olympics, right? Mostly the USA has been winning for basketball. It's just a lot less variable because it's the best of the best. It's the best of the best representing the country versus like Way more different variability in teams and rosters and changes for like individual. Okay, there we go. Let's see. Ah, come on. Wait now. Where is the verdict? What the model? Let's see. Okay, city. I mean, this makes me sad. Uh, my model, our model has predicted, this is our model and this will have like, okay, so our model predicted OKC Thunder winning it all. And second would be the Spurs. Um, it's not, well, we, our model has the Rockets fourth. And, uh, so that's a big difference. Our model got our champions, the previous 2 champions wrong. So this is really good. So again, our model, it predicted, it's, it's, it is predicting. Oklahoma City winning the title. And we are predicting as far as our top 5, we have Oklahoma City, Spurs, Knicks, Rockets, Nuggets versus— let's see what DraftKings had their top 5. It was, well, tied Spurs, OKC, tied Philly, Knicks. So basically, um, top 5 for, uh, DraftKings is Oklahoma City, San Antonio, Philly, Knicks, Boston. So what are the differences between ours and DraftKings? We have the Rockets in our top 5 and we have the Nuggets in our— we have 1, 2, 3, 4, 5, 1, 2, 3, 4, 5. We have the Rockets and the Nuggets in our top 5. Wow. All else the same, they have Philly and Boston in their top 5. So this was fun. And I mean, even despite like, what they were saying, like not having all the data, to be honest, like in terms of like the data that like Vegas uses, I mean, the fact that our model was, you know, 3 out of 5. Let's see, where is, I mean, Houston is not even, okay, Houston is, let's see, Houston, Denver. Where do they have Denver? Denver, okay. Yeah. Okay. Interesting. I mean, listen, Nuggets, they won the title not too long ago. They got their Joker. I mean, Rockets have Kevin Durant, so, and they have Fred VanVleet back. So, and they have a great, a really good coach. So, I mean, I could see, I, this, I can see, I mean, I, I can see this happening. And again, with the, with the 76ers, great on paper, but availability, injuries. That's a big thing. The great thing, the thing about OK, well Spurs very healthy with the Thunder. I know that they lost Dort, but I like this. I'm going to post this. Gonna post a link to this as well. So our model has Oklahoma City Thunder winning, and essentially I'm gonna say it looks like essentially it has Oklahoma City and the Knicks in the finals with Oklahoma City. Oklahoma City winning. So, uh, I'm gonna post this. We'll say our model lost. It did not predict the winners for the past few seasons. But again, when it comes down to predicting, you know, NBA champions, there's just so many different variables, um, way beyond data. So not necessarily, not necessarily worried about that, but the fact that it has like OKC and like Spurs and like the Knicks top 3 in terms of like odds. I think this is pretty good, pretty decent. Um, because again, it really doesn't matter if what we've predicted for like, you know, the 17th or whatever, like it's really the top, you know, these top 3 and 5 or whatever. And the winners in the past 5 seasons or so, unlike FIFA, where it's actually more like FIFA and like the Olympics, like the winners are like always like the same, like the winners for Winners for the past NBA, NBA championships. It's been, you know, quite different. It has not been like, again, look how different it is. Knicks, first time since the '70s. Oklahoma City Thunder, Boston, Denver, Golden State. It's been a different one every single time. So yeah, it's, it's completely fine. Like, I think we have a good model here. I'm going to post this. I'm going to share the data. I'm going to share this link here. This is something that, you know, you could put on your, you know, something you can put on your portfolio. Like, you get like a, you can, great thing about using Claude, you can, I'm going to make this sure any can be shared publicly and copy it. And literally it's like a webpage. Super cool. And yeah, so. A lot longer. Got the, you know, took a long time for step 1 of getting the data and like figuring out what, you know, predictors and what data we want to include. And yeah, if you do this yourself and you do anything different, let me know what you predict or what your model comes up with. And shoot, we'll come back next, you know, next summer to see if this was correct.