Machine Learning Interview Series #5-Asked In Interview ⭐ ⭐⭐⭐⭐⭐⭐⭐

preview_player
Показать описание
We at iNeuron are happy to announce multiple series of courses. Finally we are covering Big Data,
Cloud,AWS,AIops and MLops. Check out the syllabus below.


Incase of any queries you can contact the below number. Happy Learning!!

8788503778
6260726925
9538303385
8660034247
9880055539
Рекомендации по теме
Комментарии
Автор

So If I assume,
FP, in this case, is: Our model predicted that the customer would spend more than 5K, but in reality, he/she didn't, then the company's coupon may trigger him or her to buy this time, thereby adding new customers. The only drawback which I think in this case would be some coupons might get wasted as the original customer was never interested in our products.

FN, in this case, is: Our model predicted that the customer would not spend more than 5K, but in reality, he/she does spend more than that. In this case, the customer was anyway buying the products from the site.
But if he got news from his/her friends (who aren't shopaholics) that they got coupons. It may lead this customer to start purchasing from a new account or a different site.

Seems like FN is more important as original customers are getting lost, if not given benefits.
If we correctly identify them and give them the coupons and benefits. It would trigger others who are shopaholics but aren't spending much on this site, to buy even more for extra benefits.
Do let me know your view.

aviranawat
Автор

I think you don't want to lose potential customer who will be willing to spend 5000USD after getting coupon. You absolutely want to send coupons to all those potential customer who will definitely going to spend 5000 USD, so my guess is we should focus on False negative and most important feature would be users average spending monthly, average spending during coupon festivals, average spending using coupon and other buying behavior features

CRTagadiya
Автор

Must try to reduce FALSE POSITIVE.
As the company gives coupon on November itself, we can understand that company assumes or predicts that customer WILL BUY for 5000. If this prediction goes wrong 1000$ coupon will cause for a loss. So we have to focus to reduce FALSE POSITIVE. Hope my observation is correct... Welcome suggestions....🙏

hrishikeshnamboothiri.v.n
Автор

False Negative since it's a perfect example of Type 2 error i.e. "under reaction". Incorrect not rejecting the null hypothesis when it should not be rejected.
Different features could be :
1. Frequencies of the customer visit
2. Number of items purchased in each visit
3. Age group
4. Payment methods
5. Gender (M/F)
We can use Recall/Sensitivity to resolve this.
Recall = TP/(TP + FN)

deepjoshi
Автор

Reducing False Negatives because:
FP-> Customer didn't buy but predicted as bought
FN-> Customer bought but predicted as not bought
We should try to reduce FN in this case because the customer is predicted to buy but our model has predicted that he wouldn't buy, so if we try to reduce it, we can make profits by sending coupons to the more righteous customer and gain profit than to send coupons to the customers who may not buy (FP) and I believe these coupons do not add up to the loss of the company as these coupons just tend to be more of an promotional offer.
Features:
Customer age
Customer's city
Past purchases
Customer's purchasing pattern on categories
Usage of past promotional offers

ashiksrinivas
Автор

Features: participated in previous sale, amount spent in previous sale/amount spend in orders purchased throughout the year, amount of items in user wishlist according to category. I would intend to reduce false positives as the objective is to selectively choose the customers

dunnasuryanarayana
Автор

I thik it should be a business dicision whether they are ok with giving some extra coupons to non potential customers or may ok to loose some potential customers and its always better to reduce both, but if i have to strictly choose between fn or fp then i will go with reducing false negative bcz we are loosing potential consumers which is giving them 20% profits.
Features - monthly income, expenditure, frequency of visiting the site, age, last festive season expenditures, items in cart, items in wishlist

biswajitsen
Автор

False positive
features:- completed orders, current cart status, time since present on platform, %age increase or decrease in orders in festive time, how much he/she does online payment, belong to metropolitan city or not, sentiment on last 30 days orders, sentiment on last years orders etc

ravitanwar
Автор

If the coupon can only be applied against a purchase of 5k or more then fn needs to minimised, else it depends on the company completely. Potential features would be purchasing history on & off sale, wishlist, delivery location(s), order cancellation history, frequency of visit, items in 🛒 if any, customer loyalty(ratings & feedback) etc.

birupakhyananda
Автор

We should be focusing on decreasing False Negatives.
And about the feature we should be focusing on the customer sales over the year and months, does a customer uses discount's or coupons during a sale and all the basic information of a customer regarding age, state etc.

Yashgupta-ddeq
Автор

I think false positive should be reduced because if model falsely predicts customers spending more than 5000 the company will suffer loss by distribution of 1000 dollar coupons ...while if it gives false negitive results the company will not suffer any losses...
For second part the features which should be considered can be customer shopping interval, customer shopping amount, on what items does the customer usually spend more, customer shopping frequently or not etc.
If i am wrong please do correct me !

makrandrastogi
Автор

false-negative -- should be less (why - as we need to push big-bugest customer to buy by giving them coupons) and feature - brand, spending-range, last coupon used, last big buy, return, quantity, delivery location

zodiac
Автор

False negatives are those customers which the company is going to skip as they’ll be shown as not buying the products worth 5k dollars but in actuality they’re the potential customers
false positive are those customers who are not going to buy 5000 dollars worth of products and will be sent those coupons.
Now it makes sense to reduce false negatives because they’re the target customers that we don’t wanna miss and even if false positive increases and some of those customers buy 5k worth of products just to use those coupons it will be an added profit.

danielsharma
Автор

I think reducing false positive will be much more effective than reducing false negative because in false positive company is going to loose their 1000 dollar vouchers but in false negative company is not loosing anything it will be the loss of customer and company will be more look onto their loss irrespective of the customer, and the main features can be the particular type of product which is more expected to be bought in Christmas and the the geographical location, that which area has maximum probability of buying the products more than 5000

arslaneqbal
Автор

Focus on reducing False Positives.. features can be monthly salary, family income, monthly spend on groceries, purchase on discounts in past, lockdown announced or not, mode of purchase such as cash, card or online transaction, type of card used for purchase like debit or credit, credit limit, bucket of timeline analysis

vallimuthaiyah
Автор

we should try to decrease false positives because ..we have falsely predicted that they will buy more than $5000, and rewarding them with $1000 will be a loss for the company.

as far as features are concerned :
1) we could collect the customer's total expenditure during any festival (this being Christmas ) during the months from oct to jan for the past 5-10 years
2) frequent visits
3) check the customer's cart if he/she has any festival-related items that he/she could buy

these are some domain related features we could acquire
rest other basic features like gender, age could be added

sunnygeorge
Автор

😀Of course need to reduce False Positive...🔵 Then and only then company can save its coupons for right people.. False positive means your model predicted positive (will spend 500$) result which is turned out tobe negative (that person will not spend 500$) and in that case company unnecessarily would give them coupons which user wont gonna use them.. Means its overall loss to the company. Hence they should reduce FP so as chamces of issuing coupons to wrong people will be reduce and company can save its money...

gauravpatil
Автор

False negative needs to be avoided, as false positive might bring in new spend pattern.
Features: frequency of customer visit, average spent during non-christmas period using past data, average spend during christmas period, gender, age, highest spent till now, religion.

supernova
Автор

Sir after completing the Videos in the roadmap video u describe, how can i move towards job searching, which platform should i chose as a fresher.
Can u make a video on ML job searching for freshers?

dibinification
Автор

I was working on this kind of POC, basically ensuring discount for customers falling into the different clusters according to Recency, Frequency and Monetory, concept is also known as dynamic pricing.

shubhendusharma