# 统计代写|统计与机器学习作业代写Statistical and Machine Learning代考|MATH4432

## 统计代写|统计与机器学习作业代写Statistical and Machine Learning代考|Fraud in Prepaid Subscribers

Support vector machines are useful in pure prediction tasks such as fraud detection. In these cases, a full interpretation of the model outcomes is not necessary, and the main interest is in the classification results or the model decisions. A scenario like this is when a telecommunications company creates a marketing promotion to boost new cell phone operations. The promotion gave subscribers prepaid telephone minutes to make calls at the same amount as the minutes received from any telephone.
Fraudsters immediately identified an opportunity in this scenario. The fraud occurs when the criminals acquire fixed lines to connect to a computer that uses a soft switch that is a call switching node in a telecommunications network. From the soft switch, the criminals generate robocalls to hundreds of cell phones to obtain the prepaid minutes. Of course, the criminals had no intention to also pay for the fixed lines. In this case, it is a double fraud against the telecommunications company. Criminals can also use altered pay phones to generate these calls, and then the prepaid credits, or illegal extensions to switches to do the same, to make calls to prepaid cell phones and generate credits to them. Once these prepaid cell phones are full of credits, the criminals sell these prepaid phones on the black market. In many markets, there is no need to provide any information to the carrier when purchasing prepaid subscriber identification module (SIM cards).

The main goal of the fraud detection model is to detect unusual behavior in receiving calls and block the prepaid minutes credit. Some useful variables include no active behavior in making calls, exceedingly high and consistent behavior in receiving calls, very narrow list of originating numbers, and originating numbers at the same place, sometimes at the same address. The SVM model classifies the cases based on a binary target, 0 or 1 , or in this case, fraud or non-fraud. This classification is based on the posterior probability of the event, or the likelihood of a prepaid fraud. Based on the posterior probability, the prepaid phone numbers with the highest posterior probability are sent to a deny list that cancels their prepaid credits. Fraud detection models can save telecommunications timely monitor the events. They must be closely monitored to assess their performance to predict fraud cases. They are often updated frequently, as fraud events change often. For all these reasons, the main goal for fraud detection models is the performance, rather than the interpretability, of the model. This is a perfect case for neural networks, random forests, gradient boosting, and support vector machines.

## 统计代写|统计与机器学习作业代写Statistical and Machine Learning代考|Factorization Machines

There is a class of business applications that involves estimating how users would rate some items. In this business scenario, very often companies do not have much information about the users nor about the items. The main goal in this type of model is to evaluate relationships between users and items. For example, consider a streaming based company selling on-demand movies to users. This company has little information about its customers (users) except the movies (items) they watch. Users can eventually rate the movies that they watch. If we create a matrix considering users and movies, considering the combination of a rating given by a user to a movie, this final matrix will be very sparse. We can expect, for example, millions of users and hundreds of thousands of movies. Not all customers rate all movies. In fact, few customers rate movies, which leads to a huge matrix comprising a substantial number of missing ratings. In other words, most of the cells in that matrix would be missing. The challenge here is to estimate what would be a user’s rating to a particular movie.
Factorization machines are a new and powerful tool for modeling high-dimensional and sparse data. A sparse matrix is the matrix of users and items of the movie company we just mentioned. The main goal of factorization machines is to predict the missing entries in the matrix. By estimating the missing entries, the company would know all ratings for all movies, considering all users. That information would allow the company to find items that users would give high ratings and then recommend those items to each user (except the ones who have rated those items).
Factorization machine models are used in recommender systems where the aim is to predict user ratings on items. There are two major types of recommender systems. One type relies on the content filtering method. Content filtering assumes that a set of information about the items is available for the model estimation. This set of information is often referred to as side information, and it describes the items that will be recommended. By having the side information available, the content filtering has no problem in estimating the rating for a new item.
The second type is based on collaborative filtering. This method does not require additional information about the items. The information needed is the matrix containing the set of users, the set of items, and the ratings of users to items. The combination of all users and items creates a large, sparse matrix containing lots of missing ratings. The collaborative filtering method works well in estimating ratings for all combinations of users and items. The problem with the collaborative filtering method is the inability to estimate added items that have no ratings.

Figure 5.4 shows an example of collaborative filtering. In this case, all the information the company has is the existing ratings, that can be represented in a matrix. Not all the users rate all the items, so some of the ratings will be missing.
Factorization machines estimate the ratings by summing the average rating over all users and items, the average ratings given by a user, the average ratings given to an item, and a pairwise interaction term that accounts for the affinity between a user and an item. The affinity between users and items is modeled as the inner product between two vectors of features: one for the user and another for the item. These features are collectively known as factors.

# 统计与机器学习代考

## 统计代写|统计与机器学习作业代写统计和机器学习代考|分解机器

myassignments-help数学代考价格说明

1、客户需提供物理代考的网址，相关账户，以及课程名称，Textbook等相关资料~客服会根据作业数量和持续时间给您定价~使收费透明，让您清楚的知道您的钱花在什么地方。

2、数学代写一般每篇报价约为600—1000rmb，费用根据持续时间、周作业量、成绩要求有所浮动(持续时间越长约便宜、周作业量越多约贵、成绩要求越高越贵)，报价后价格觉得合适，可以先付一周的款，我们帮你试做，满意后再继续，遇到Fail全额退款。

3、myassignments-help公司所有MATH作业代写服务支持付半款，全款，周付款，周付款一方面方便大家查阅自己的分数，一方面也方便大家资金周转，注意:每周固定周一时先预付下周的定金，不付定金不予继续做。物理代写一次性付清打9.5折。

Math作业代写、数学代写常见问题

myassignments-help擅长领域包含但不是全部: