This research addresses the challenge of managing Key Account Managers (KAMs) by introducing and validating the “7 KAMs Journey” framework, a structured lifecycle model spanning from candidate selection to retirement. Moving beyond subjective evaluations, this study employs and compares two machine learning algorithms, XGBoost and Support Vector Regression (SVR), to analyze integrated historical data from HRIS and CRM systems, aiming to predict performance and identify dominant success factors at each career stage. The empirical results demonstrate that a KAM's career trajectory can be accurately modeled, with XGBoost showing a slightly better overall predictive fit ($\mathbf{R}^{\mathbf{2}} \boldsymbol{=} \mathbf{0. 5 5 8 0}$) compared to SVR $\left(R^{2}=0.5387\right)$. The models achieved particularly high accuracy in early-stage predictions, with XGBoost excelling at forecasting onboarding success ($\mathbf{R}^{\mathbf{2}}$ =0.9358) and SVR strongly predicting long-term performance from candidate data ($\mathbf{R}^{\mathbf{2}} \boldsymbol{=} \mathbf{0. 8 0 3 2}$). Key findings reveal that quantifiable attributes, specifically standardized assessments (CnQ Test) and formal certifications, are powerful predictors of future success. This study contributes a validated, data-driven framework for KAM management, providing organizations with a strategic roadmap to optimize recruitment and development using robust predictive models, ultimately empowering them to cultivate a high-performing KAM workforce with greater precision and foresight.

Network-layer ransomware attacks are on the rise with the rapid deployment of Internet of Things (IoT) devices, particularly in environments with limited CPU, memory, and energy resources. Conventional defenses are often too complex, allowing stealthy ransomware traffic to bypass protection and compromise service reliability. This paper introduces a lightweight hybrid framework using Network Simulator version 3 (NS-3), which combines simulation-based rule enforcement with machine learning analysis. The rule-based mechanisms include token-bucket rate limiting, circuit-breaker quarantine, authentication validation, replay protection, and Fair Queuing Controlled Delay (FQ-CoDel). These mechanisms reduce excessive packet injection, quarantine malicious nodes, and sustain fairness across IoT flows. However, some ransomware traffic still evades filtering. To improve resilience, supervised learning classifiers, Random Forest, Support Vector Machine, and K-Nearest Neighbors, are trained on controller node logs. Results show that FQ-CoDel stabilizes throughput at approximately 45 kilobytes per second and maintains an average drop rate of 52 %. Random Forest provides the best detection accuracy of 84 % with an area under the curve of 94 %, while K-Nearest Neighbors records 79 %, and Support Vector Machine achieves 77 %. These findings confirm that integrating lightweight rule-based controls with machine learning enhances ransomware protection in resource-constrained IoT system.

Indonesia's oil palm business has become a valuable commodity as palm tree plantations have expanded. On the other hand, the technology deployed needs further transformation to improve the effectiveness of industry processes. As part of the engineering process, dilution attempts to separate oil from sludge during the clarifying process, which requires 30 percents water to be added to crude palm oil to get the best product quality. In mathematical terms, the controlled plant is a nonminimum phase model that is difficult to control. Our research seeks to create a control system to govern the percentage value of diluent water using PID control, with the best PID parameters obtained utilizing the particle swarm optimization (PSO) approach. The PSO algorithm found that $\text{Kp}=1.435$ and $\text{Ki}=2.560$ resulted in steady feedback and rapid reaction, with tp (peak time) = 0.75 seconds, ts (settling time) = 1.40 seconds, Mp (maximum overshoot $)=13$ percents, and $\operatorname{tr}$ (rise time $)=0.50$ seconds.

Accurate and efficient water meter reading is a crucial aspect of modern utility management, directly affecting billing accuracy and operational transparency. However, manual data collection remains widely used in many regional water companies, often leading to high labor costs, human errors, and data delays. This paper proposes a vision-based smart water meter reading system to support utility digitization through an integrated image signal processing pipeline that combines object detection, computer vision enhancement, and optical character recognition. In the first stage, a You Only Look Once (YOLO)-based detection model automatically identifies the Region of Interest (ROI) containing the numerical display of mechanical water meters under varying lighting and orientation conditions. The second stage performs preprocessing, including grayscale conversion, noise suppression, contrast normalization, and deskewing, to improve character clarity. In the final stage, Tesseract Optical Character Recognition (OCR) extracts numerical readings from the enhanced image and converts them into structured digital data. Experimental results demonstrate that the proposed method achieves a precision of 0.97, recall of 0.95, mAP@50 of 96.4%, and mAP@50–95 of 94.7%, with an inference time of 0.18 seconds per frame. These outcomes confirm the framework’s reliability, scalability, and real-time capability for smart water meter reading in utility digitization.

Slot Backdoor attacks have become a growing cybersecurity threat in Indonesia, particularly exploiting vulnerabilities to inject unauthorized online gambling advertisements into university websites. These attacks alter website appearances, disrupt academic services, and negatively impact Webometrics rankings by reducing site accessibility and credibility. University XYZ experienced a severe decline in its Webometrics ranking, dropping from the top 5 to around 1000 due to repeated Slot Backdoor attacks. Following the deployment of the proposed machine learning-based mitigation system, the university successfully recovered and returned to the top 5 rankings. To address such threats, this study proposes an anomaly detection and prevention system at the network layer using machine learning. The system integrates Isolation Forest for anomaly detection and the Random Forest for attack classification. Isolation Forest identifies users with more than four failed login attempts within an hour, triggering a warning. If no further attempts follow in the next hour, the status returns to Normal; otherwise, access is Blocked to prevent brute-force and backdoor attacks. Random Forest then classifies network anomalies into Normal, Distributed Denial of Service (DDoS), and Malware/Backdoor threats. Evaluation metrics accuracy, precision, recall, F1-score, and confusion matrix confirm the system's effectiveness. The Random Forest classifier achieved 93 percent accuracy, with 88 percent precision for DDoS, 97 percent for Malware or Backdoor, and 100 percent for Normal traffic. Isolation Forest demonstrated 100 percent accuracy in DDoS and malware/backdoor mitigation with minimal false positives. By combining anomaly detection and threat classification, this approach significantly enhances cybersecurity resilience and offers a practical solution for protecting university networks like University XYZ.

Weather disaster such as floods and strong winds frequently affects various regions in Indonesia, including Gebangan Village in Grobogan Regency. Due to difficult terrain and vast area of the village, an Internet of Things (IoT)-based early detection and warning system was developed to improve community preparedness and response to such disasters. The system consists of three main subsystems: monitoring nodes equipped with sensors i.e. anemometer, wind vane, pressure sensor, rain gauge and river level detector; gateway node that receives signals from monitoring nodes using LoRa communication and the MQTT protocol; and web-based monitoring system that displays real-time data and automatically sends WhatsApp notifications when critical conditions are detected. Testing results show that the system can reliably transmit data between nodes up to 1 km, accurately read environmental parameters, and deliver timely alerts. This integrated system enables communities to receive real-time environmental information to support rapid and effective disaster mitigation decision-making.

In the competitive landscape of Business-toBusiness (B2B) markets, customer loyalty plays a strategic role in sustaining revenue growth. The Net Promoter Score (NPS) is a popular loyalty metric, yet it often falls short in reflecting actual customer behavior, particularly when high NPS scores do not translate into repeat purchases or sales growth. This study introduces a machine learning-based approach to enhance NPS evaluation by integrating it with customer behavior metrics using the Recency, Frequency, and Monetary (RFM) framework. A dataset comprising NPS and transaction records was preprocessed and analyzed using the Random Forest algorithm. The integration of NPS and RFM enabled the identification of customer segments that conventional NPS analysis might overlook, such as high-scoring but lowcontributing customers. The findings reveal that combining behavioral insights with loyalty metrics leads to more accurate customer profiling and actionable strategies for retention. This model supports data-driven decision-making and provides B2B firms with a more reliable method to align loyalty indicators with financial outcomes.

Face recognition has advanced significantly due to deep learning, primarily via libraries like Dlib. This paper proposes adaptive upsampling in the Residual Network (ResNet)-34 deep learning architecture to improve face detection in low-resolution images. The Support Vector Machine (SVM) is integrated into the fully connected layers classification tasks to improve accuracy. Principal Component Analysis (PCA) reduces feature dimensions, reducing training time and increasing detection rate. ResNet-34 in Dlib is trained using 128 face feature vectors and Euclidean distance on the Labeled Faces in the Wild (LFW) dataset. Reducing PCA dimensions from $\mathbf{1 0 0}$ to $\mathbf{5 0}$ or SVM cost parameters from 100 to 50 has no noticeable impact on accuracy. Still, it gives more Frame Per Second (FPS) in real-time recognition because the parameters are fewer.

Tourism significantly contributes to the economic development of regions like Central Java, Indonesia. Therefore, innovative approaches are essential to enhance the tourist experience and provide destination recommendations that align with visitor preferences. Realizing this, researchers introduced an integrated model that combines rule-based sentiment analysis with a decision support system of the Simple Additive Weighting (SAW) method. We apply SAW to refine the recommendation system for choosing tourist destinations. This methodology uses online reviews of tourists for 10 destinations in Central Java. Using Aspect-based sentiment analysis, researchers categorize the sentiment expressed by tourists into 3A aspects, often referred to as Attractions, Amenities, Accessibility based on predetermined keywords related to the three main aspects of tourism. The results of the sum of criterion 3A and these sentiment subcriteria are then fed into the decision support system of the SAW method using Attraction, Amenity, Accessibility as the main criteria and positive, negative, and neutral sentiments as subcriteria. This integrated approach combines feedback from online traveler reviews based on the 3A concept with the SAW method providing a balanced and accurate recommendation system. The implementation of this integrated model demonstrates its good ability to generate user-focused destination recommendations. Where Lawangsewu is ranked number one with the highest weight, namely 0.5242 obtained from the weighting of the SAW decision support system method. Consistency in this ranking was tested by changing the weight by 5%, and after testing, the ranking order of tourism destinations did not change. Showing the potential of this research approach can be used as a reliable measuring tool in helping to recommend the selection of tourist attractions in Central Java.

Monitoring the water quality of shrimp ponds using conventional methods involves periodic water sampling manually performed by pond farmers from various locations within specified intervals. This method is not only time-consuming but also susceptible to uncertainties, as rapid changes in water quality may not be effectively detected. Hence, a machine learning system with a Random Forest algorithm is proposed. The system can monitor water quality in real-time, continuously, and automatically. Moreover, it can process realtime data and classify the water quality of shrimp ponds based on predefined variables against the standard values established in the Indonesian National Standards (SNI). The water quality classification process involves comparing pH, temperature, and TDS features with the SNI. The research results indicate that the Random Forest classification model achieved accuracy and precision of 94.91 percent and 97.44 percent, respectively. These performances are better than those of KNN and SVM. Moreover, the most influential feature of the output model is TDS, with a value of 50.7 percent.

The inefficiency of traditional warehouse systems can lead to a 20% decrease in revenue and a 30% reduction in storage capacity within five years. Therefore, this study aims to design an AGV (Automated Guided Vehicle) capable of picking up and storing assets in the warehouse. The design focuses on enabling the AGV to choose the most optimal route. To achieve this, the method used in this research involves creating a warehouse map model using graph theory and employing the A* and D* algorithms for route selection. The heuristic value in both algorithms is calculated using Euclidean distance in addition to AGV turning costs. The findings demonstrate that, within the confines of the developed warehouse model, the D* algorithm perform slightly better than A* algorithm by approximately 1.02%. These results offer valuable insights into the efficacy of the D* algorithm in conjunction with graph theory mapping for warehouse navigation. By revealing the slight performance advantage of D* over A* in a static warehouse environment, this research contributes valuable insights for Industry 4.0 implementation. These findings can inform the selection of optimal pathfinding algorithms in warehouse automation, ultimately leading to improved efficiency and competitiveness within Indonesia's logistics sector.

Our research presents a novel approach to enhance the recognition of individuals wearing masks, which can be particularly challenging due to obscured facial features. By incorporating our method into the ResNet-34 deep learning framework, we aim to overcome difficulties in detecting masked faces in specific frames. Typically, a mask complicates facial recognition, as it merges all features, including the mask itself, into a singular facial representation. We used a selective feature extraction strategy to address this, focusing on specific face landmarks. We tested 5, 25, and 31-point landmarks within the ResNet-34 model, adapting our method to improve masked face detection. Additionally, we utilized an adaptive technique and refined our feature selection process, supporting the enhancement of detection rates by focusing on smaller subsets of facial landmarks when a face is not initially recognized. Further accuracy improvements were achieved by employing a Support Vector Machine (SVM) in the model's final layers, providing a refined approach to feature differentiation without relying on principal component analysis (PCA) due to the mask's influence. We also explored variations in ensemble model structure, finding an optimal configuration for face recognition involving a shallower cascade regressor. Our best results were obtained with settings of 31 landmark points, a 15- depth cascade, and a tree depth of 4, successfully identifying 300 out of 359 masked faces.

One of the potential implementations of mobile robots in mitigating the risk of communicable diseases is for delivering items and handling materials with no-human contact. In that scenario, mobile robots require the capability of performing simultaneous multi-object detection and distance estimation for dynamic exploration in a real-world environment. Some existing research papers have contributed to the development of object detection or distance estimation, separately. However, there is still a lack of studies on simultaneous multi-object detection and distance estimation for robot navigation, especially in terms of performance analysis of the integrated techniques. One of the most challenging issues of integrating multi-object detection and distance estimation is the validation and evaluation of the type of objects with their relevant estimated distance to the robot. For that reason, this research aims to analyze the mobile robot's sensing performance in integrated simultaneous multi-object detection and distance estimation during navigation scenarios. A depth camera of Intel RealSense D455 and a rotating LIDAR of Slamtec RPLidar A1M8 are utilized in the experiments. The result of experiments shows that the object detection algorithm using Mobilenet SSD V2 can perform an accuracy of 74%, with precision, recall, and F1-Score consecutively achieving 74,54%, 74,00%, and 74,22%. The multi-object distance estimator also performs satisfactory results with a low measurement average error of 10.3 cm for indoor experiments. The integrated multiobject detector and distance estimator system can work in a real-time mode of 30 FPS, thence significantly improving the navigation performance of the mobile robot.

Dalam tugas akhir ini yang berjudul Sistem Informasi Evaluasi OBE Prodi dan Pengukuran CPL Mahasiswa. Judul tugas akhir ini berawal dari riset bersama dengan dosen yang dilatarbelakangi dari Aspek penilaian seringkali menjadi hal yang dianggap sebelah mata sehingga untuk mencapai target yang diharapkan akan menjadi semakin sulit. Maka dari itu diperlukan sebuah inovasi berupa sistem baru yang dapat mengakomodir kebutuhan berdasarkan tantangan yang ada. Sistem yang akan memantik mahasiswa untuk lebih eksploratif dalam meningkatkan kemampuannya, menghadirkan kelebihan serta kekurangan kegiatan pembelajaran secara komprehensif, serta evaluasi terhadap hasil yang didapatkan secara konkrit. Outcome-Based Education (OBE) adalah sistematika pendidikan yang berpusat pada keluaran atau outcome, bukan hanya materi yang harus diselesaikan.

This research proposes a practical solution for seamlessly integrating PHP with Python in web development, focusing on achieving efficient web crawling. The problem is that many PHP applications need to call the Python application for machine learning or crawling work. With default functions from PHP, such as PHP-exec, the Python program could be executed but cannot be maintained smoothly if the program is a lengthy task in the background. By leveraging the AMQP (Advanced Message Queuing Protocol) library and the Selenium Crawler, Celery, and RabbitMQ, we establish interoperability between PHP and Python. In our approach, PHP acts as the front end, initiating web crawling tasks by doing some action in the web component. These requests are queued with a message broker application such as RabbitMQ, the message broker. RabbitMQ connected with Celery for the seamless scheduling and execution of tasks. This research enables effective web crawling and concurrent data scraping by seamlessly integrating the Selenium Crawler with Celery. Results from extracted data from the crawler are saved to the database to give a certain status of whether the data collection process is done or pending. Through experimentation, we validate the effectiveness of our seamlessly integrated approach by making a variation of worker and concurrent connection. The testing scenario shows that increasing workers only increase a small amount of memory. This result indicates that workers could help maintain the response time if there is some user, but need some consideration based on the number of users and availability of memory and CPU.

Ad-Hoc Networks (VANET) merupakan pengembangan teknologi jaringan wireless dari MANET. VANET terdiri dari kelompok kendaraan bergerak yang dihubungkan oleh jaringan nirkabel. Hal ini memungkinkan mobil mengirim dan menerima informasi satu sama lain. Penggunaan utama VANET adalah untuk memberikan keamanan dan kenyamanan bagi pengemudi khususnya untuk kendaraan mobil. Dalam beberapa tahun ini terdapat peningkatan jumlah kendaraan otonom khususnya mobil, sehingga kebutuhan akan VANET menjadi sangat penting. Jaringan VANET dalam mengirimkan control packet menggunakan broadcast routing dan tidak semua broadcast routing mengirimkan data paket. Broadcast routing ini dapat menimbulkan permasalahan berupa Broadcast Storm Problem (BSP). Hal ini yang menjadi dasar dalam melakukan penelitian mengenai evaluasi kinerja perutean broadcast routing pada VANET. Kinerja dari broadcast routing pada VANET dapat dinilai berdasarkan metrik secara umum seperti average end-to-end delay yang menunjukkan rata-rata waktu yang dibutuhkan sebuah paket untuk bergerak pada jalur dari node sumber ke node tujuan dan packet delivery ratio yang digambarkan sebagai rasio paket data yang diterima di tujuan akhir dengan total paket yang dihasilkan oleh sumbernya. Penelitian ini direncanakan dengan simulasi dengan algoritma no underlying topology khususnya yang menggunakan protokol Fixed Transmission Range (FTR) dengan penilaian kinerja average end-to- end delay dan packet delivery ratio dengan variasi pada jumlah node yang disimulasikan pada perangkat lunak Simulasi Urban Mobility.

Agriculture is an important economic sector in an agricultural country, such as Indonesia. In developing agricultural potential, there are many modern agricultural systems, one of which is the greenhouse system. The greenhouse is a planting system that can prevent plants from several factors that can cause damage to growth. Greenhouses interact with the surrounding environment and create a microclimate in the greenhouse. Changes in the microclimate are very influential for greenhouse cultivation plants. This paper provides a design of a two-way data communication system based on the Internet of Things (IoT) for monitoring and controlling smart greenhouses. Smart nodes can send microclimate readings to a remote database every 15 minutes. Users can control the actuators in the smart greenhouse remotely. Analysis and testing of the design of a smart greenhouse-based two-way data communication system were carried out by calculating the latency and packet loss data values. The average response latency of remote actuator control is 14.83 seconds. In sending microclimate data, smart greenhouses have an average latency of 14 minutes and 56 seconds with a packet loss percentage of 0%.

Rumah pintar merupakan sebutan untuk rumah yang memiliki berbagai perangkat yang dapat terhubung satu sama lain menggunakan koneksi internet untuk berkomunikasi. Dengan kata lain, kita dapat memantau perangkat atau keadaan rumah/bangunan tersebut dari jarak jauh. Beberapa hal sederhana, seperti menyala matikan lampu dapat di atur dari jarak jauh. Namun system smarthome yang umum digunakan masih sangat bergantung pada internet, dan tidak dapat berkerja saat internet tidak ada. Tujuan perancangan system smarthome local network ini untuk membuat system pada smarthome masih dapat menjalankan perintah sederhana walaupun tanpa koneksi internet. Hal itu dapat dicapai dengan mengadakan jaringan local dengan mengandalkan koneksi Wi-Fi tanpa internet atau disebut Local Wi-Fi pada system rumah pintar, yang membuat perangkat dapat bertukar data secara lokal tanpa harus terhubung ke internet. Dari pengujian , kualitas pengiriman data yaitu waktu tunda pengiriman sangat dipengaruhi oleh jarak dan ada tidaknya hambatan. Pada kondisi dengan hambatan dan tanpa hambatan, semakin jauh jarak modul ESP 01 maka waktu tunda semakin besar.

Temperature is an important factor in growing plants up. It affects the transpiration, photosynthesis, metabolic rate, and respiration of plants. Water spinach (Ipomoea Aquatica) can grow optimal at a temperature of about 25°C-30 °C. In the greenhouse, the increase of air temperature can cause the plants die. In this study, Levenberg-Marquardt's method of artificial neural networks is used to predict the temperature inside of the greenhouse 30 minutes in the future. The system utilizes inside-outside humidity, light intensity, and outside temperature of the greenhouse as the input. The training yields the correlation coefficients of training, validation, and testing with the value of 0.99. These values are utilized to produce a prediction model. Furthermore, a mist maker and fan use the model to decrease the temperature. In addition, Levenberg-Marquardt's method was compared to the fuzzy method in terms of suppressing the temperature. The result of temperature range inside the greenhouse are in range 25 ± 1,48°C until 30 ± 0,16°C and 25 ± 2,5°C until 30 ± 8,7°C for the former and the latter, respectively.

In the landing process, airplanes have many safety factors that must be protected. Weather advice and information is very important as a consideration in determining the feasibility of landing an aircraft. The main objective of this research is the implementation of Mamdani fuzzy simulation in determining the feasibility of landing aircraft at Ahmad Yani Airport in Semarang using ATC police and pilot coordination on runway number 31. Wind direction, wind velocity, visibility, and pilot experience are used to determine eligibility aircraft landing. An intelligent system based on fuzzy logic produces three decisions that are feasible, careful, and not feasible in landing an aircraft on a runway. The results of the study concluded from an intelligent system based on fuzzy logic can be used to determine aircraft landing decisions on the runway of Ahmad Yani Airport Semarang on runway number 31.