 Before I start, I would like to make a couple of disclaimer. After listening to so much machine learning, deep learning, and data mining, I'm going to talk about the topic which you feel totally away from the data mining. I call myself as an analyst, I call myself as a data scientist, I've been involved more than 15 years in data analytics, but what I'm presenting is not what I have learned, it's about connected vehicle. So, and also I'm working for SAS for more than a decade. I'm not representing SAS from the connected vehicle point of it. It's more like from last two years, I'm working as a consultant and whatever my experience and understanding about when I'm dealing with the especially the manufacturing as a client, how I solve the problem and to see kind of a end-to-end perspective, how you see the connected vehicle. That's what I'm going to present. But nevertheless as an expert, I'm here to answer whatever I know about, what kind of analytics we can do offline, kind of online analytics, whatever the kind of the questions. Now, we do have a lot of time, so do stop me anytime whenever you have questions and I analysis it. So, before we start with what the connected vehicle is very important that we understand what the internet of things is. So, most of my talk is focusing on what it is, how we capture the data, how the sensors are put once we get the data, how important it is to compress the data before we do the analytics. So, a lot of the focus is on IOT and then couple of the application beyond connected vehicle that I'm going to talk. It's not only connected vehicle as such, how we can use as a smart city or in a different manufacturer. Getting into the troubles of copyright or patents or any kind of mechanism of control. The analytics or how we capture the data and analyzing the connected car and then open for the question answers. So, when we talk about internet of things, I hope you can see the details. So, it's not only just capturing the data and getting the insight for a specific domain. It goes across the domains. It's starting from transportation, communication, smart cities, the healthcare and it's going. Common Crawl is a non-profit organization that built and they've constantly been maintained. There are kind of analytics on the top of the data and then trying to answer the questions what the client is looking for. So, when we think of IOT, the main promise that IOT can make is that the efficiency. How efficiently we are going to answer the problems that we are looking into and the quality of the life. When we say quality of the life, can I identify a false or a defect much before it happens? Can I answer how we can stop the defect or a failure of any part of a vehicle for example? Can I give the alerts? For example, we are talking about so many issues in the school bus. Can I have the sensors in the school bus so that the parents can get the continuous alerts about what's happening? Is it possible that before anything happens that the parents will kind of reach to the driver or reach the authority and kind of solve the problem? So it's not only when we say connected car is not only to capture as much as data but also to filter the data in a way that we get our right data and do the analytics on the top of it and get the insights. Do know that we are not talking about a big models here. I'm not talking about RNN, I'm not talking about probably random forest. At this point of a time, it's not important to come up with a big model, improve the accuracy as such. We are not here to say I got the 99.81. It's not kind of a research paper where I want to improve the accuracy. It's very important to give the alert at the right time so that the action has been taken. So I just want to switch with all the prior talks that it's kind of different all the game here what we are talking about. So when we talk about the IoT promise, the main three things comes here, the big data. When we again talk about the big data, it's not only capturing, keeping the sensors, it can be related to the vehicle, it can be related to the smart city. It's not just keeping the sensor and capturing this data, it goes beyond about it. IoT data plus probably beyond the IoT data. When I say beyond the IoT data, it might be the customer related data, might be a personal information, might be where the individual is located into, might be the product information. So all together that data need to be taken to identify how the insights can be captured and the alerts or whatever we want to identify the failure of a vehicle or issues in a smart city can be captured and the action can be taken well before something goes well, something goes bad as such. So then comes on the top of the data what analytics we can do. Yeah, I mean, I'm assuming all of us including all of you are having a lot of knowledge about all the models in this world, right? So I'm repeating this as if I'm talking about a layman, I'm not talking about a data scientist here because I know there is one person I can go and reach out what are the best models to handle this data. My question is whether this model answers this data or not. I mean, model answers the questions that I'm looking out of this data or not. If they answers my end client, is it going to understand or not? Because we did talk about interpretability and all, that I'll come soon. So at this point of a time, I'm not talking much about how good my model is. Is it sufficiently good to answer the question that I'm looking forward? And then comes once I have the analytics, what value addition I can do and how efficient I can get the answer. Can I improve the quality of the life? Can I have the early warnings as such? And what are the best business model that I can think of to improve whatever solution at this point of time I have? So all this is of no use unless I act on it. Is there any use I'll come up and say, okay, whatever the sensor data I got, I'm able to find the fault. I'm driving a car. Oh, I got to know that my engine is going to break. Is it sufficient? No, I want to act on it. I would rather have someone recognize that the engine is going to fail and take an action, at least point me to the nearest service station or at least point me to say stop there, do not drive more. So I want some action on the top of it. Then only it makes sense to have rather than running at that time the best model and try to resolve that. I would rather know how action can be taken so that I know that that failure doesn't happen and it has been taken care as such. So the best, I mean, the part is unless you act, there is no value for the analytics or the data that we have captured. So in simple words, what I'm trying to say is that the entire IOT, which essentially talks about is the science, you need to sense what all the data you have from variety of sensors. And then you understand essentially building the model so that you can capture the, you know, every part of the data so that you can answer the end questions which the user is looking for and act with whatever the model you build so that you know that this is going to solve the problem. So the main challenges here is the first part when we think about is the data. So when we are collecting data, again I'll talk about, you know, how much data we get especially in connected vehicles. So is it all the data is going to be useful? Most of us know that independent of whatever the, you know, problem we are trying to solve that, the data whatever we have is of not completely useful. We might want to look exactly what the data we have to look for. But here the other challenges, we have so much data, we don't know whether we want to use everything in case if you want to use how you store it and it's more like a real time problem, right? As the data is coming, you need to understand can I use the data, you have to analyze the data on the, you know, online, on the real time and you have to capture, can I store all the data or I have to capture only a part of the data? If I have to capture the part of a data, am I going to repent at the end that oh, I missed the data, you know, I lost it? So all this decision has to be well made so you know that exactly what data you have to retain so that that will help you to take the decision at the end point of the time. And you have to keep on monitoring as the data is coming and do note that independent of how good the sensors are, you know, the data, the quality might not remain the same throughout the time. So you have to analyze at each point that the quality of the data, how good the data is coming and added to that, is it the quality data that you want to capture for the end question that you are looking forward? Is it going to help your model to answer what you're looking for? Then comes, you know, this is the traditional way of solving the problem, right? That what we have seen, we do have the data, you know, we kind of store the data and on the top of it we build the model and then depending upon the, you know, probably the predictive modeling part of it or you know, whatever the kind of the model we build, we generate an alerts and looking at the alerts we kind of, you know, take an action for that. So how different our IoT analytics is going to be? It's very much same, nothing much different. This has been happening from last 20, 30, whatever the number of years. So the additional change is going to be the IoT. So the IoT data when we are talking about is not only the census data, the data is coming from the social media, the data is coming from, you know, different part of it, as I said, might be from the products, the data might be from the customers, it might be from the blogs, it might be from any other sources than the sensors as such. So the end number of, you know, the ways in which the data is coming in. So on that particular data, the biggest challenge is to do the intelligent filter or to kind of the transform the data. So throughout connected vehicle or the IoT analytics, more than 80% the value addition is how you filter intelligently so that the data what you got is going to give the best answer to the question that you're looking for. I keep on saying the questions you're looking for because it's going to different from domain to domain, problem to problem. Here, it's not that I got the best data on which I run the model and I got the best prediction. It's not the way in which it's working. You have to filter it, you know, differently or a different, I should say, intelligently so that it exactly answers the model what you're looking for and hence answers your question as such. So then comes streaming model execution. Again, when we are seeing streaming model, it's not the offline, right? It's not the data is there, historical data and you build a model. There is quite a bit of part, you know, historical data analysis there as a part of IoT. But what I'm talking about is as the data is coming in, you have to calculate a lot of transformation. Might be it's a simple average or it might be simple standard division or something but when you're doing it has to be quite meaningful. So what kind of analytics you do when the data is streaming, you know, on the point, it's not taking the data and do the historical. So that's much value addition to it. So once you have, so depending upon what you transformation have done, you can generate alert as simple as that. If your speed or the temperature in the coolant or the temperature in the engine beyond this generate an alert, that is sufficient. You don't have to build a big model to answer the question here. So any simple analytics, which is done on the real time, which will generate an alert and helps you either to control the force or to help the driver or to do any changes which will help the performance of a vehicle, that's a big value addition here. So just, you know, so here if you see what the CEO of Ford has said, it's, I mean, the successful companies, you know, are not only going to be the ones that provide the best product, but also the ones that collect the best data and capture the best analytics for at the end to answer what exactly the questions they're looking for again. So here, the connected vehicle, it's essentially enabling a lot and a lot of business opportunities as such. So few of them here, you can see one is, you know, the car sharing and the loan sharing, the smart mobility and multimodality. You can also, there's one pay for miles drawn and the data-driven insurance. So all of these are a different opportunity. Just, you know, with the simple model techniques, you can try to answer only the challenging here is how well you capture the data and how well you can filter the data in simple words. And there's another thing, I mean, most of us, I'm sure in Bangalore, how bad the traffic is, most of our life is spending car, right? So what all you want to do in the car? I mean, you might want to, I mean, one thing is the automated car so that, you know, you are all free to do whatever you want. Probably you just, you know, checking or, you know, you kind of check into the office as soon as you enter the car and the car is yours, right? For the internet, for the kind of the entertainment or might be, you know, you want to work or you want to do kind of social chatting, whatever it is. All these things are possible when you're a car, but nevertheless, all this data has to be, you know, captured offline. So anything is coming in, you should know, you are, you know, working in a car, you are entertaining yourself in the car, probably in whatever, two, three hours of your, you know, travel, during that time, whatever happens will be captured along with your personal information. When I say your personal information, it can be, you know, what all blocks you entered in, it might be what you responded and I'm not talking only about the census in the vehicle, about the individual as a driver or as a, you know, passenger in the car. So that depending upon your, you know, mental status or a psychological or behavioral information, one can say, okay, it's probably it's not right time to go to office, take a turn and, you know, take a break and then go, you know. It is to that extent that one can capture all the possible information and analyze and kind of guide you. And, you know, one of the clients that I'm working on, they were more interested just to let the, you know, the driver know how early they can get to know they're tired, right? The second thing they wanted to know, is it possible for them to tell them that the gas is going to be over and what is the nearest gas station? If not, is there any other way so that the driver can be replaced? You know, they're very, very simple questions for which we don't have to, you know, think too much about what the model part has to be and not to worry about too much, the machine learning or a deep learning kind of a techniques, but to kind of think from bottom up, you know, saying that, okay, this is the question to be answered. What are the sensors that I need to put on in the car and which data I need to look and, you know, what kind of a data I need to filter which is coming out of the sensor and do a very simple analytics which can solve the problem. So, overall, I mean, there are, I mean, the recent, 2018, one of the papers online analytics, how random forest can, you know, do the best predictive modeling and the whole agenda or the whole, you know, challenging here is not which model to use, then the question will come, why only the random forest? And as a researcher, you'd rather, you know, try every technique or find out a new algorithm to solve a problem, but here what I'm stressing again and again here is that we are not here to solve the problem, you know, using the best technique, but to solve the problem in a better way so that we get a right answer. So here, you know, a few more things that, you know, what are the opportunities as such we can think of along with what we talked just before. And from the business perspective, you know, essentially that we want to optimize the business, you know, we want to generate the revenue and customer has to be happy at the end of whatever we kind of do the analytics. These are the kind of the end goals that you can focus on when you are trying to solve a problem. So one important thing here is how, you know, safety the data what we are looking at the client is because it's very important that, you know, the data is safe because so much data is coming we might say that we want to discard the data because it's not that important, but nevertheless throughout the journey of, you know, capturing the insight, it's very important to have a safety of the data. So when we're trying to, you know, get a kind of a interaction with the customer, what all are the things that we can look for? One is certainly the safety and security as I mentioned and how the customer is experiencing at the end of the solution and the quality and the reliability and the infotainment that I just talked about and the dealer operation and also the location-based services. For example, might be for that when we are talking it can't be a generic solution. Might be for that specific location these are the one which makes more meaningful. So it can be the customized solution when we are talking about. So again within a safety and security we can think of, you know, the accidental risk reduction and a reduced risk what can be the cyber security we can think of capturing. When we are talking about a location-based services contextual driving services. So as I talked about, you know, when we are talking about the driver behavioral analysis probably when we want to ask him to go to a, you know, nearest service station probably we can analyze is the contextual kind of a data and we can kind of a persuade him, right? This is the right place to go rather than going to the other place, right? And talking about dealer operation, the streamline operations, parts inventory optimization we can think of and the network optimization as such. So the, yeah, please. Here is that the, as I'm saying the, you know, again it's my understanding at this point is that when we are talking about us safety or security is that much before a fault happens we are going to tell the individual that, you know, we are reducing the possibility of the risk. So the accident is going to happen. I have identified the whatever the failure. So I got an alert. So I'm taking a precaution of addressing that alert and hence reducing the, you know, the risk of the accident, for example. That's one example that you can think of. So in the car. So the essentially it is that with whatever the data I, we would rather generate an alert which will address the accidental risk much before it happens from that angle. I'll go a little bit quickly so that we can go for the further one. So here the quality opportunity. So here it essentially would like to say that, you know, the one is when we're looking for the quality opportunity asset, one is the under utilization, the scrap or the rework or the warranty cost or the lost sales. So these things we can, you know, focus on and we can think of addressing much before, you know, it's again what we talked about the, you know, getting before it happens, we can take care of the lost sales if we kind of capture the data. Again, here if you say lost revenue, you can think of customer churn cost. If you think of excess inventory, carrying costs. So these few sales can be much improved. Again, if you capture the right data from the sensor and address it. The warranty cost again, or even the sale penalties. Again, if you, you know, capture the data and much before you focus on, you know, the, already the products are released and you get to know that they're faulty. So before calling upon all of them, you would rather say that this is a part of the vehicle which has failed. I would rather address only part of it so that, you know, essentially that particular cost has been taken rather than calling of the entire slot which has gone for the production. So here, so mainly here the focus is, as I said, the early warning. As soon as we find the defect, we will kind of address it. So here most of the companies rely mainly on the warranty claims, might be from the caller center or any other resources to provide the early, you know, indication of any issue with the problem. So instead of depending on that, if you just do the, you know, kind of the entire analysis and find out any possible failure of any part of the vehicle and essentially generate, you know, identification of this failure, then most of these cost has been taken care of. And then just beyond what the claims, what we can think of. Okay, one is the, with respect to the business drivers, it's reduced cost, especially the automotive early warnings, accelerate the root cost analysis, you know, why this happened, can we resolve this and minimize the recall size, as I mentioned, increase the customer satisfaction, fix it, but at the right time. It's not that the customer comes back and say that I'm having this problem. So we'll identify, you know, again, depending on these, you can decide where the sensors has to be placed. For example, most oftenly, failure parts, for example, probably I'll focus the sensors, data from those sensors, rather than the generic data across the sensors. So the point of focus can be depending upon what the historic kind of the, you know, complaints or the defects were coming. Accordingly, the data can be captured and analyzed so that you can, you know, essentially minimize the recall size, for example. And then enhance, I mean, this will help you to improve the brand image as such. So the main challenge is to move beyond the traditional data sources, so which are like usually the warranty, the call centers and, you know, the telematics data, for example, it essentially will, you know, capture every part of the vehicle so that you don't have to worry about, you know, kind of the traditional way. These are the only ones that which I need to worry about. So this telematics data, you can just analyze it and you can capture every part of it and understand, okay, because of this, the issue is coming and if I focus only on this, I'll generate an alert accordingly and, you know, solve the problem as such. Yeah, the same thing. So here, what are the new opportunities as such? So predictive maintenance models can leverage the raw sensor data to anticipate any individual failure mode. So we are just talking about our raw data. You still haven't gone a level up to, you know, recognize can we do more analytics on that. Then next is detecting these failures, you know, again in the raw data for any specific car, you know, would support the development of a new service model itself, which is going to again help to the, improve the customer value as such. Models, so one step ahead is generating any models or building any models on the top of this data, certainly is going to remove, you know, further eliminate latency in the issues that going to, you know, kind of deletion of the entire process as such. So the entire lifecycle that we talked about is, you know, can we improve the entire process by deploying the right model? So here, if you can see that, you know, the entire process that we talked about, data, you know, starting from ad hoc analytics that we talked about, and then we take a decision, what kind of analytics we need to do, and kind of building a model, and then data storage. Again, here is the right decision to make, you know, what kind of a data can be filtered, and what level data we can keep in, and then streaming model execution, during the data streaming, what kind of a models I can build in, those model, how they are going to help with respect to the offline analytics, where I can, you know, historically collect the data and build the model offline, and then, you know, deploy those models to see, you know, how better my decisions are with respect to the online analytics as such. So the value to the IOT, couple of the things to quote here, that fever products are shipped with the problems resulting in a very low warranty cost, you know, 10% to 15%, and then new service models that derive value to the customers and efficiency throughout the service work, improved customer satisfaction and the customer retention, and again, as I said before, it will keep the band name kind of untouched. So when we are talking about our onboard data processing, so essentially we, you know, acquire the data kind of the transfer, we do a different transformation as such, and during this again, it's very important that how we compress the data, do we keep only the compressed data, you know, stored, or do we keep the stream, streamlined data, or do we do the analytics on streamline, and do we do a different analytics on the transform data? So these are the major decision that we need to take, again, depending upon what problem kind of we are trying to solve. Transmission to the ground instrument and the sensor data. So online analytics, you can see that, you know, just to give couple of examples here, the four connected car sensors, almost 25 gigabytes data only for hour or so, and Wall Street Journal quoted that typical autonomous vehicle generates four terabytes of data within 19 minutes, and Intel, you know, for about 45 terabytes per hour. So it's not all small data that we are tackling with, right? And so when we are talking about our data, so what all kind of a sensor or what all kind of a data we can think of capturing across the vehicle as such. So when we are talking about a vehicle location, it is the GPS coordinates, it's a speed limit, it's a accelerometer kind of information, it's a compass orientation. Each of this means a difference in, you know, kind of the analytics. Most of these might be, you know, if you just see the example that I take, just looking at the accelerometer kind of the observation, it might be simple just to observe it and any kind of outlier is going to give you information here. And in fact, in most of the sensors, once you capture the data and you decide, you know, kind of filter, any outlier is information here. It's, I mean, probably in the real time, you have to address them separately, but once you kind of a transform and, you know, understand it, the noise and the outlier is going to add a lot of value when you do the online analytics as such. So these are the small things which are very specific when you are trying to capture, you know, during the stream data, during, you know, online analytics and when you do the offline analytics. Perka. Driver train metrics. So this is, again, you know, the big challenge, for example, Olbo and, you know, Aishur, so their main thing is focused on the driver train metrics. So the driver's status, the engine RPM, engine temperature, fuel level and fault codes. I'm talking about only the main one, which is useful, but when we keep this and so it's much beyond that. And when we're trying to capture, when we go and say that, you know, as a expert you get, go and say, when I want to capture any information about a driver or the driver behavior analysis, it's good for me to have, for example, you know, what highlighted like a driver status or engine temperature or, you know, fault codes. It's not easy that I just go and keep at the specific location and I get the sensors for just to capture, you know, six or seven of them might be, I'll get a hundred different variables. So from hundred to come for only this size is the big challenge. So what I'm trying to say is that it's very important, it's not even simple variable reduction, what I, what we talk about. It's more to understand out of hundred measurements, which essentially capture the driver train metrics, but what I might, it's not that out of hundred only 30 are important, okay? It's not that, you know? So it's very important to understand where the sensors needs to be kept. And once after keeping the sensors, what data is going to be useful? And within that data, what it's worth storing so that I'm not, you know, misusing the space and also I'm not making my models tired by running with unwanted data, for example. So these decisions are much more important than thinking of the best model to solve a problem, you know, connected vehicle, I can say. When we are talking about the vehicle, environmental status, it's like what's the temperature within the cabin and what is the, you know, the outside kind of a temperature? Can we detect the rain? Can we identify the humidity around? And the customer sensors, as you know, the cameras, you know, tracking services different and the temperature across the location and what is the damaging impact on a client as such. So scalable, big data management analytics for the connected cars work in a totally different way with respect to the verticals that we discussed before. It's just very focused on kind of fault detection to understand, can I, you know, understand the behavioral of a driver so that I can kind of persuade him to take the right decision at the right time because, you know, it's very important to understand when drivers, especially the loaded trucks, the ISHER, the problem is, it's very hard for them to tell the driver they are tried because they want to earn more, right? So it's very hard for them to persuade him, right? This is the kind of the limit that you can drive. So is there any way to tell them that this is, you are done, you know, but you are tired, this is the time for you to take a break? Probably when the next load comes, you can drive. This is the time and by the time the next driver is ready to kind of continue the journey. So to understand those things, you need to give them the alert to the driver and persuade the drivers saying that you're really tired, right? So these are very simple things for that. It's not that you just take the data, analyze and give the decision. The alerts has to be generated quickly and along with the knowing the alert, you have to persuade the driver that it is the right time for you to take the decision. Otherwise, you know, things might not work or passenger or the load might be in trouble. So I just categorize, you know, this is a kind of a high level kind of a classification. Usually we talk about online or offline analytics. So here it's like at the edge of analytics, in stream analytics, at rest analytics. So when I said the edge analytics is during the process when it's streaming on the spot, the analytics kind of done. For example, this thermostat kind of a sensor with processing capability. It works with a minimal context to the data, confide to very rudimentary rules as such. And a very simple statistics as average standard deviation, simple statistics. Simple commands can be automated in this case, just like, you know, thermostat adjust or any variation is going to give an alert, for example. In stream analytics, it takes place when data streams on device to the other. Analytics on multiple stream inputs has a richer context and can be used to identify the much more complex pattern as such. A simple example that you are aware of is mobile phone use relative to, you know, subscribe plan can be triggered based on the location and activity. It's just an example. So at rest analytics is the core what we have been talking, you know, today and last evening, that is the historical repository data. You can do any analytics the best that you can think of, you know, it's more rich and, you know, there is no limit for it here that you can work on this. So here we talk about creating any best predictive models and, you know, kind of forecast and whatever the new patterns that you can think of, you can capture using here. So this is more like offline analytics. So I don't want you to talk about your descriptive predictive and all, but I just wanted to give example because most of us think, oh, descriptor is just a descriptive. We are not here to talk about descriptive. So I just wanted to highlight a simple example how it can be used with respect to connected car because even those simple things add a lot of value here. So here, movement by movement, driving patterns or road behavior of all, you know, kind of individual who get behind the wheel of a particular vehicle. And these patterns can be evaluated and linked to a safe over risky behaviors. And these can be combined with other data sources in the vehicle information. So when we do it, it might be stand alone. This might not have any value, but together we'll give a lot of value as you see further. So again here, a very simple example that, you know, is driver is going to likely is going to take different way, a shortcut than the best path from the road condition point of right. So very small, those things, probably we kind of a do a thing and then combined with a descriptive analytics and say, okay, so that's better than just stand alone descriptive analytics or stand alone predictive analytics as such. So again here, the example that I wanted to quote was that, for example, coaching guidance to specify where an individual should start slowing down, okay, to safety navigate a curve on the road or guidance to ensure that, you know, the vehicle is being well-mounted at convenient location. So I mean, when you think of it, you don't have to think of a big model. It's a very simple things you can do and you can kind of guide it. What is the right way to do at this point? So here, one of the, just the flow to understand, you know, kind of when the trouble code appears, what is the right way or what is the simplistic way to solve? So the predictive analytics can recognize that a driver normally drives between two places on a typical route and detects that a driver is on the different course. Very simple path, if you just go across, that's going to help a lot. You know, it might can be simple prescriptive, even that's going to have a value addition. So here, this is what I wanted to highlight with, you know, there were a couple of the queries that came to me. Can you tell us when we can run a random forest? Can we run our decision-free or a random forest, which works best? You know, at this point of time, with whatever limited experience that I have, it's not the random forest, it's not the decision-free, it's not the RNN. What you need to understand at this point of it, is it interpretable? Especially because most of these models are going to be executed in a batch board, probably on the kind of a historic and on a real time. And the client want to understand, it's not that you deploy the model and the model's giving with whatever the accuracy forever. So in this case, interpretability is very important. Every point the user wants to understand why they got the result and does it making sense and to retrieve back or point of, go back to the problem and understand is it going to give the right answer or not? Certainly the computation, the efficiency and nevertheless the accuracy is the main, but the main point here, what they're looking for is the interpretability. So just wanted to highlight here, what exactly they're looking for. For example, might be it's very hard. There are a lot of papers, how you interpret to the common man about random forest, for example. But nevertheless, if I'll just explain, for example, in a simple here, we had built both the models and when we try to explain to them that this model is giving the best results, it was more like, usually you know that, right? We work with the domain people, right? To understand the problem requirement and the one who deploy the model at the client side is the ITT, right? So the IT people are answerable to the domain people. So the first question to them is, can I understand the random forest? For me, the accuracy, whatever, the 99.9 is not that important. Even I come close to it is fine. What is easy for me to interpret to the end user, right? So again, I'm trying to say, for example, when we compare technically the decision-trialed random forest, I mean, might be the random forest is better at one point and you know, might be decision-triusing, there are a lot of pros for that also. But at this point of the time, I would like to say the decision-tree makes sense because it was more easily interpretable and it's very easy to consume for them as is. Might be it is in one case, but it's very important at every point you understand that rather than having a complicated model, very simple model which can answer might not be the perfect answer but which can relatively answer. So here to just give the example, you know, what kind of data you are, you know, getting what's the size of the data that you are going to capture from the car? So if you say here, again, it's from the Intel, though, every kind of a driver car, you know, will churn about almost 4,000 GB, okay? Again, it is per day that we are talking about, okay? And I mean, it was compared with the, you know, individual average persons, the kind of the data, what the access with the social media and the video which turns out about just a 650 MB, you know, per day. So you can compare, you know, what's kind of the data that we are getting in. So one reason for, you know, cars appetite is the hundreds and lots and lots of sensors. What I was talking about is just to capture, you know, kind of the diverse thing. We had hundred sensors. So there are so much sensors, accordingly you'll get lot and, you know, kind of a lot of data. So yeah, just added the quote from the, you know, the Intel where they're trying to say each car driving on the road will generate about as much as data as the 3000 people, you know, and just a million, you know, autonomous vehicle you can think of will generate three billion people's worth of data as such. So I'm just taking couple of the examples for the IoT. So it's not only the connected vehicle. There are many more examples that I mentioned. The one is the connected factory. So here, essentially identifying the, you know, hidden patterns that essentially predict the, you know, failures of improving the production yield as such and connected service. So here, operation through the predictive maintenance, you know, how you kind of control it. So it essentially, you know, provides the manufacturers with new insights and proactively, you know, identify the equipment issues and how you can address them before the client comes and, you know, tells what's the problem. And, you know, already about the smart cities and the connected energy again here can we improve the, you know, kind of a profitability and kind of a customer service by leveraging the data from the smart meters as such. So connected car, we already talked about there's a connected customer, you know, essentially to, you know, kind of predict what are the kind of customer preferences, right? And again, it's in the real time, right? Time we're giving the suggestions and, you know, oftenly give the offers which customers immediately are ready to offer, for example. The connected health, connected transportation, and running out of time. So I'll just go ahead. Connected auto insurance, connected gaming and connected customer as such. So just few examples to quote here, you know, many, many, you know, already the automobile are already on to it. The old one, I should together already started in India for the school bus, connected school bus. Already the project is up and running for them. And even the I should trucks, already they have started, you know, connected vehicle trucks as well as for the buses. Again, few of the examples to talk about. So just want to have a time for, so I'll just run a small demo here. I'm not sure. So just, I mean, I'll not be able to, you know, so this is kind of a real time, you know, it recognizes kind of the, any change in the speed or, you know, kind of a remotely monitoring. And as soon as the change, it goes and, you know, the dashboard, what the driver is looking is going to show and what is, you know, creating a failure of a vehicle or a failure of engine. Driver's dashboard will itself will highlight so that what the driver is known. And accordingly, the driver will get a message how it can be resolved. So it kind of a navigate that these many kilometers your service center is that, you know, the fuel is reducing, this is the closest service gas station is or you stop the car and there is a support probably on the highways where there is no gas station. There is a supporting, you know, kind of the supporting gas we are providing within whatever the limited time. So these things are, I'll just go to the final one. Just it's kind of a captures the, you know, the vibrations. So as soon as you get to see the which vehicle has gone beyond the specific, you know, the vibration, you get to see a kind of alarm event, you go with the text thing, what has made it and also accordingly what the action needs to be taken for example. Yeah. So the general, I'll just show you the combined thing, which Stop here. Any questions? I'm sorry. The questions need to be taken offline for this session. Yeah, I'm around and you know, I asked the focus was not only the analytics to cover, so I haven't focused on any analytics but I'm open for any kind of a conversation. I'll be around here. Feel free to contact me. Thank you so much.