Collection Prediction and Homogeneous Assignment System with Machine Learning

Abstract

The invention relates to a system that optimizes the processes of estimating the collection risk of indebted customer files in the telecommunications sector and, based on these estimations, equitably and homogeneously assigning files to external sources (law firms).

Description

1 TARIFF Collection Prediction and Homogeneous Assignment System with Machine Learning Technical Area 5 The invention enables estimating the collection risk of outstanding customer files in the telecommunications sector. And based on these forecast results, a fair distribution of services to external sources (law firms) with a system that enables the optimization of homogeneous file assignment processes It is related. 10 State of the Art Telecommunications companies need timely payments from their customers to survive. They are obliged to collect the unpaid bills. These collections must be completed by 15. For this purpose, they generally work with external resources (law firms). In the current process, the company, legally Companies refer clients who are under investigation to law firms. On a monthly basis, firms report legal matters to law firms. They determine their performance based on the amount their offices manage to collect, and this It determines the number of files to be assigned the following month based on performance. In the current process, The main problem preventing fair and transparent performance evaluation of offices is 20 The problem is that the files assigned to the offices are not of homogeneous difficulty. If the assigned clients are equal If they are not in a difficult situation, the offices cannot be evaluated fairly. For example, some offices are in debt. While many customers with low amounts were assigned, some had high debt levels. Customers can be assigned. This leads to transparency issues and the company's achievements It risks reducing the amount of revenue it will collect. Today, there are 25 ways to solve these problems. Although various methods have been developed, the existing solutions remain insufficient, therefore the relevant technical... An improvement in this area has become necessary. Purpose of the Invention The invention was created by drawing inspiration from existing situations and overcoming the aforementioned drawbacks. It aims to solve the problem. The main purpose of the invention is to reduce the potential for customers who are subject to legal proceedings to pay their debts or to estimate the probability using machine learning techniques and this estimated probability 35 The goal is to assign clients of equal difficulty (homogeneous) to offices using their values. 2 Another aim of the invention is to make the debt collection process in the telecommunications sector fair and efficient. The aim is to ensure that it is brought to that state. Another aim of the invention is to increase the transparency of the collection process and the total amount collected. to increase. 5 The structural and characteristic features and all the advantages of the invention are given in the figures below. Thanks to the detailed explanation written with references to the figures, it becomes clearer. This will be understood, and therefore the evaluation should also take these figures and detailed explanations into account. It must be done by taking 10. Figures that will help understand the invention. Figure 1 shows the general architecture of the system that is the subject of the invention. Description of Part References 15 1. Raw customer dataset 2. Data preprocessing module 3. Feature Engineering Module 4. Feature selection algorithm 5. Machine learning prediction module 20 6. Probability of Payment Score 7. Homogeneous assignment algorithm Detailed Description of the Invention In this detailed description, the preferred configurations of the system that is the subject of the invention are listed only. This will contribute to a better understanding of the subject and will not have any limiting effects. The invention enables estimating the collection risk of outstanding customer files in the telecommunications sector. And based on these forecast results, a fair distribution of services to external sources (law firms) It is a system that enables the optimization of homogeneous file assignment processes, 30 Initial data provided by the telecommunications company for customers who have been subject to legal proceedings. raw customer dataset (1) containing data in raw customer dataset (1) cleaning, filling in empty values, categorizing (mapping), and Data preprocessing module (2) which performs normalization operations, new data from existing data Feature engineering, which enables increasing the number of features by deriving descriptive features. 35 module (3) is the optimal subset that provides the highest information gain in the dataset. 3 The feature selection algorithm (4) that enables the determination of the customer's using a trained model. machine learning prediction module (5) which estimates the likelihood of paying off debt, machine The learning estimation module (5) produces and shows the potential for debt repayment. The payout probability score (6), which is a numerical output, will be assigned using the payout probability scores (6). The average difficulty (average probability of payment) of client groups (by law firm) is 5. It includes a homogeneous assignment algorithm (7) which ensures that it is homogeneous. The operating principle of the system described in the invention is as follows: The system described in the invention is telecommunications. It is based on the principle of making the debt collection process fair and efficient in the sector. For this purpose, the raw customer dataset (1) and attribute information, data preprocessing module (2) and attribute 10 The data obtained as a result of these preliminary preparations are processed by the engineering module (3). Using the feature selection algorithm (4), the optimal feature sub-selection algorithm that will give the most accurate result The set is determined. This optimal subset is trained on the machine learning prediction module (5). and a payment probability score (6) is calculated, which shows the probability that a customer will pay their debt. The basic working principle of the system is to use these payment probability points (6). These points are 15 then evaluated by the homogeneous assignment algorithm (7) and assigned to law firms This ensures that the average payment probability of the customer files to be sent is equalized. In this way, the transparency of the collection process and the total amount collected are increased. The machine learning prediction module (5) used in the system that is the subject of the invention predicts the customer's payment 20 A number of supervised learning algorithms were used to predict its behavior. These among them Logistic Regression (LR), Decision Tree (DT), Naive Bayes, K-nearest neighbor (K- EYK), Artificial Neural Networks (ANN) and Ensemble Methods (Adaboost, Bagging, Gradient) Boosting, Extra Gradient Boosting (XGB) is available. The feature selection algorithm (4) used in the system in question improves the prediction performance. In order to increase and shorten model runtime, tree-based methods (average) values), Principal Component Analysis (PCA), Recursive Logistic Regression and Chi-square Different feature selection methods, such as tests, have been used. The feature engineering module (3) used in the system that is the subject of the invention is telecommunications. Data obtained from the company, including attributes (subscriber city, subscription duration, gender, debt, etc.). Cleaning, reduction, and expansion processes are performed.

Claims

4 REQUESTS 1. Estimating the collection risk of indebted customer files in the telecommunications sector and Based on these forecast results, outsourcing (law firms) will be done fairly. It is a system that enables the optimization of homogeneous file assignment processes, 5 feature; • Initial data provided by the telecommunications company for customers who have been subject to legal proceedings. raw customer data set containing data (1), • Cleaning the data in the raw customer dataset (1), filling in empty values, Data pre-processing that performs categorization (mapping) and normalization operations. processing module (2), • Increasing the number of features by deriving new descriptive features from existing data. feature engineering module (3), • Identifying the optimal subset that provides the highest information gain in the dataset. feature selection algorithm (4), 15 • A machine that predicts a customer's likelihood of paying their debt using a trained model. learning prediction module (5), • The machine learning prediction module (5) produces and the potential for debt repayment The payout probability score, which is the numerical output showing (6), • Using the payment probability scores (6), assign client groups (law firm 20 homogeneous assignment that ensures average difficulty (average payout probability) (based on) algorithm (7) It includes.

Images